_ Since 1955 ^ Strojniški vestnik Journal of Mechanical Engineering Strojniški vestnik - Journal of Mechanical Engineering (SV-JME) Aim and Scope The international journal publishes original and (mini)review articles covering the concepts of materials science, mechanics, kinematics, thermodynamics, energy and environment, mechatronics and robotics, fluid mechanics, tribology, cybernetics, industrial engineering and structural analysis. The journal follows new trends and progress proven practice in the mechanical engineering and also in the closely related sciences as are electrical, civil and process engineering, medicine, microbiology, ecology, agriculture, transport systems, aviation, and others, thus creating a unique forum for interdisciplinary or multidisciplinary dialogue. The international conferences selected papers are welcome for publishing as a special issue of SV-JME with invited co-editor(s). Editor in Chief Vincenc Butala University of Ljubljana, Faculty of Mechanical Engineering, Slovenia Technical Editor Pika Škraba University of Ljubljana, Faculty of Mechanical Engineering, Slovenia Founding Editor Bojan Kraut University of Ljubljana, Faculty of Mechanical Engineering, Slovenia Editorial Office University of Ljubljana, Faculty of Mechanical Engineering SV-JME, Aškerčeva 6, SI-1000 Ljubljana, Slovenia Phone: 386 (0)1 4771 137 Fax: 386 (0)1 2518 567 info@sv-jme.eu, http://www.sv-jme.eu Print: Grafex, d.o.o., printed in 380 copies Founders and Publishers University of Ljubljana, Faculty of Mechanical Engineering, Slovenia University of Maribor, Faculty of Mechanical Engineering, Slovenia Association of Mechanical Engineers of Slovenia Chamber of Commerce and Industry of Slovenia, Metal Processing Industry Association President of Publishing Council Branko Širok University of Ljubljana, Faculty of Mechanical Engineering, Slovenia Vice-President of Publishing Council Jože Balič University of Maribor, Faculty of Mechanical Engineering, Slovenia Cover: Hollow parts are applied in a variety of structures, from automotive and aircraft designs to household appliances. When solid parts such as drive axes and shafts are replaced with their hollow counterparts, the weight of mechanical structures is greatly reduced. Lower vehicle or aircraft weight, in turn, means a higher power of the engine and reduced fuel consumption. The interest of the automotive and aircraft sectors in hollow parts is due to rigorous emission standards for engines. New manufacturing techniques are being researched and developed. Courtesy: Lublin University of Technology, Faculty of Mechanical Engineering, Poland ISSN 0039-2480 International Editorial Board Kamil Arslan, Karabuk University, Turkey Josep M. Bergada, Politechnical University of Catalonia, Spain Anton Bergant, Litostroj Power, Slovenia Miha Boltežar, UL, Faculty of Mechanical Engineering, Slovenia Franci Čuš, UM, Faculty of Mechanical Engineering, Slovenia Anselmo Eduardo Diniz, State University of Campinas, Brazil Igor Emri, UL, Faculty of Mechanical Engineering, Slovenia Imre Felde, Obuda University, Faculty of Informatics, Hungary Janez Grum, UL, Faculty of Mechanical Engineering, Slovenia Imre Horvath, Delft University of Technology, The Netherlands Aleš Hribernik, UM, Faculty of Mechanical Engineering, Slovenia Soichi Ibaraki, Kyoto University, Department of Micro Eng., Japan Julius Kaplunov, Brunel University, West London, UK Iyas Khader, Fraunhofer Institute for Mechanics of Materials, Germany Jernej Klemenc, UL, Faculty of Mechanical Engineering, Slovenia Milan Kljajin, J.J. Strossmayer University of Osijek, Croatia Janez Kušar, UL, Faculty of Mechanical Engineering, Slovenia Gorazd Lojen, UM, Faculty of Mechanical Engineering, Slovenia Thomas Lübben, University of Bremen, Germany Janez Možina, UL, Faculty of Mechanical Engineering, Slovenia George K. Nikas, KADMOS Engineering, UK José L. Ocana, Technical University of Madrid, Spain Miroslav Plančak, University of Novi Sad, Serbia Vladimir Popović, University of Belgrade, Faculty of Mech. Eng., Serbia Franci Pušavec, UL, Faculty of Mechanical Engineering, Slovenia Bernd Sauer, University of Kaiserlautern, Germany Rudolph J. Scavuzzo, University of Akron, USA Arkady Voloshin, Lehigh University, Bethlehem, USA General information Strojniški vestnik - Journal of Mechanical Engineering is published in 11 issues per year (July and August is a double issue). Institutional prices include print & online access: institutional subscription price and foreign subscription €100,00 (the price of a single issue is €10,00); general public subscription and student subscription €50,00 (the price of a single issue is €5,00). Prices are exclusive of tax. Delivery is included in the price. The recipient is responsible for paying any import duties or taxes. Legal title passes to the customer on dispatch by our distributor. Single issues from current and recent volumes are available at the current single-issue price. To order the journal, please complete the form on our website. For submissions, subscriptions and all other information please visit: http://en.sv-jme.eu/. You can advertise on the inner and outer side of the back cover of the journal. The authors of the published papers are invited to send photos or pictures with short explanation for cover content. We would like to thank the reviewers who have taken part in the peerreview process. © 2015 Strojniški vestnik - Journal of Mechanical Engineering. All rights reserved. SV-JME is indexed / abstracted in: SCI-Expanded, Compendex, Inspec, ProQuest-CSA, SCOPUS, TEMA. The list of the remaining bases, in which SV-JME is indexed, is available on the website. The journal is subsidized by Slovenian Research Agency. Strojniški vestnik - Journal of Mechanical Engineering is available on http://www.sv-jme.eu, where you access also to papers' supplements, such as simulations, etc. Contents Strojniški vestnik - Journal of Mechanical Engineering volume 61, (2015), number 3 Ljubljana, March 2015 ISSN 0039-2480 Published monthly Papers Janusz Tomczak, Tomasz Bulzak, Zbigniew Pater: The Effect of Billet Wall Thickness on the Rotary Compression Process for Hollow Parts 149 Lei Wan, Dazhong Wang, Yayun Gao: Investigations on the Effects of Different Tool Edge Geometries in the Finite Element Simulation of Machining 157 Nenad Kapor, Momcilo Milinovic, Olivera Jeremic, Dalibor Petrovic: Deterministic Mathematical Modelling of Platform Performance Degradation in Cyclic Operation Regimes 167 Vikas Pare, Geeta Agnihotri, Chimata Krishna: Selection of Optimum Process Parameters in High Speed CNC End-Milling of Composite Materials Using Meta Heuristic Techniques -a Comparative Study 176 Zulkiffli Abd Kadir, Saiful Amri Mazlan, Hairi Zamzuri, Khisbullah Hudha, Noor Hafizah Amer: Adaptive Fuzzy-PI Control for Active Front Steering System of Armoured Vehicles: Outer Loop Control Design for Firing On The Move System 187 Zaipeng Man, Fan Ding, Chuan Ding, Shuo Liu: Study of an Energy Regeneration System with Accumulator for Hydraulic Impulse Testing Equipment 196 Pavel Ponomarev, Tatiana Minav, Rafael Äman, Lauri Luostarinen: Integrated Electro-Hydraulic Machine with Self-Cooling Possibilities for Non-Road Mobile Machinery 207 Strojniški vestnik - Journal of Mechanical Engineering 61(2015)3,149-156 © 2014 Journal of Mechanical Engineering. All rights reserved. D0l:10.5545/sv-jme.2014.1977 Original Scientific Paper The Effect of Billet Wall Thickness on the Rotary Compression Process for Hollow Parts Janusz Tomczak* - Tomasz Bulzak - Zbigniew Pater Lublin University of Technology, Poland This paper presents the numerical and experimental results of investigating the effect of geometrical parameters of tubular specimens on the rotary compression process for hollow axisymmetric shafts with central necking, and their accuracy. The numerical investigation was performed using the finite element method (FEM) with the Simufact Forming simulation software suite. The simulations investigated the effect of tubular specimen wall thickness on the kinematics of metal flow and the geometry of produced shafts. The results were validated in experimental tests using a forging machine designed specifically for this purpose. Keywords: rotary compression, hollow parts, finite element method, experiment Highlights • A new method for producing hollow parts is presented. • The relationship between the wall thickness of a tubular specimen and rotary compression is investigated. • The effect of wall thickness on the rotary compression process is studied both numerically and experimentally. • The distributions of effective strain and temperature are determined by the finite element method. • The variations in length and thickness of the work piece wall are determined, too. 0 INTRODUCTION Hollow parts, such as stepped axes, shafts, tubes, rings and many other axisymmetric parts, have a variety of applications, from automotive and aircraft structures to household appliances [1]. Due to their being highly in demand, hollow parts with finishing allowances are increasingly produced by metal-forming processes [2]. Compared to machining, metal-forming methods for producing hollow parts ensure reduced material consumption as well as enhanced strength properties of parts (fibre continuity is maintained). It should, however, be noted that the majority of current metal-forming processes for hollow parts are significantly complicated [3]. As a result, it is expensive to run them and, hence, to implement new technologies [4]. For this reason, new highly-efficient methods for forming hollow parts are being developed to ensure the production of parts with the required geometry and strength at the lowest costs possible. The innovative methods for forming hollow parts from tubular billets include hydroforming and the RoProFlex method developed by the Institute of Forming Technology and Lightweight Construction (IUL) in Dortmund, Germany. The hydroforming technique is used to produce hollow complex-shaped parts that cannot be produced as monolithic components by other techniques. Parts produced by hydroforming are applied in automotive and aircraft industries. The main disadvantages of hydroforming include the high costs of tools, the slow speed of the process, and a lack of information about the process and tool design [5]. The RoProFlex method can be used to form any cross sections of tubes and sections along their longitudinal axis. Thanks to the numerical control of tool motion, this method can be used to produce parts with limitless variations of their cross sectional geometry. The RoProFlex method is mainly applied to produce light car parts, precise elements such as gears and helical rotors, as well as medical implants [6]. Rotary compression can be used to produce axisymmetric hollow shafts, including toothed shafts with straight, helical and herringbone teeth, as well as worms. In contrast to hydroforming and RoProFlex, this technique can be employed to form parts with thicker walls. The machine for rotary compression has a very simple design. The disadvantage of rotary compression is that this technique cannot be used to produce non-axisymmetric parts that can, however, be produced by hydroforming and RoProFlex. One of the innovative methods for forming axisymmetric hollow parts is rotary compression shown in Fig. 1. Rotary compression is a metal-forming operation based on the technique of cross rolling with three rolls, in which the tools not only rotate, but they also simultaneously travel towards the axis of the work piece. In this process, sections of commercial tubes or sleeves are rotated and deformed to reduce the diameter of steps. Compared to current techniques for manufacturing hollow shafts, rotary compression offers numerous benefits, including lower forming forces, reduced material consumption, increased strength properties of produced parts, enhanced process efficiency, lower implementation and production costs, as well as simple design of the process and the possibility of its automation. Given the above benefits, it can be claimed that rotary compression can be applied in the industry. Therefore, the present investigation undertakes both theoretical and experimental analyses to develop relationships between the individual parameters of the rotary compression process for hollow parts and product quality. Fig. 1. Schematic diagram of the rotary compression process for hollow shafts; a) start of the process, and b) end of the process; 1, 2 and 3 working rolls, 4 specimen, 5 finished part 1 FEM MODELLING OF THE ROTARY COMPRESSION PROCESS The present investigation mainly deals with presenting the effect of wall thickness of tubular specimens on rotary compression and product quality in the course of forming hollow necking in the centre of a shaft. The geometry of the work piece is determined by the reduction ratio (S=D/ d) of the step (where: D is the outer diameter of the tubular specimen, d is the outer diameter of the step), which was set to 1.5 and of the initial wall thickness to of the tubular specimen set to 3, 5, 7, 9 and 11 mm, respectively. The shape and dimensions of the tubular specimen and of the desired final product are shown in Fig. 2. First, the rotary compression process for a hollow shaft was modelled numerically. The computations were performed by the finite element method (FEM) using the Simufact Forming simulation software suite version 12.0, which the authors had previously used to investigate rotary processes for forming metals. The geometrical model of the process developed for the purposes of the numerical analysis was identical to the schematic design of rotary compression shown in Fig. 1. The model consisted of three identical forming rolls and a tubular specimen (a C45 steel tube that had an outer diameter, D, of 42.4 mm, a wall thickness, to of 3, 5, 7, 9, 11 mm and a length, L, of 120 mm), modelled by means of first-order eight-node hexahedral elements. The model of C45 steel used in the computations was taken from the library database of the software used and is defined by: = 2859.85 x e(-0.00312548xT ) (-0.000027256x7+0.000818308) ) xs (0.000044662x7-0.101268) , xe xs (0.000151151x7-0.00274856) (1) where T is the temperature (ranging from 700 to 1250 °C), e is the effective strain, e is the strain rate. At the initial stage of the process, the tubular specimen was heated throughout its volume to a temperature of 1150 °C, while the temperature of the tools was maintained constant at 100 °C. In the experiments, the tools were heated to a temperature of 100 °C with a gas burner. The temperature of the tools was monitored using a Flir T425 thermal imaging camera. The rolls were rotated in the same direction at a constant velocity n = 36 rpm, and simultaneously travelled relative to the axis of the work piece at a constant velocity v = 2.0 mm/s. The friction factor m on the material-tool contact was set to 1 [7], the material-tools heat transfer coefficient was 30 kW/m2K [8], while the heat transfer coefficient between the material and the environment was set to 0.35 kW/m2K [9]. s be observed. This results from higher resistances to flow of the material during the forming of tubular specimens with thicker walls. Fig. 2. Shape and dimensions of a) tubular specimen, b) hollow shaft During the process, the rotating tools form necking in the centre of the shaft, which is accompanied by rapid metal flow radial to the axis of the work piece, thus leading to an increase in the thickness of the walls in this region. Furthermore, it can be noted that the material travels along the axis of the tubular specimen, which results in an increase in the work piece length. At the final stage of the process (i.e. sizing), the tools only perform a rotational motion, removing inaccuracies produced in the previous stages of the process. The FEM-determined geometry of hollow shafts and the effective strains for tubular specimens with different initial thickness of the wall are illustrated in Fig. 3. It is characteristic of the strains that they are distributed in a non-uniform manner. In all examined cases, the highest strains are concentrated on the outer surface of the work piece wall. The closer it is to the internal wall, the smaller the strains are. Such a distribution of strains is typical of rotary metal-forming processes and can be attributed to the kinematics of tool and work piece motion. Considerable slipping between the material being formed and the tools, as a result of which additional deformations are generated in the circumferential direction (by friction forces) on the work piece surface, can be observed. That increasing the initial thickness of the tubular specimen wall not only leads to an increase in the maximum strains, but to a greater non-uniformity of these strains can also Fig. 3. FEM-calculated distributions of effective strains in the final stage of the process for tubular specimens with various wall thicknesses The heat capacity of the hollow tubular specimens is much lower than that of the solid tubular specimens, which means that the material cools faster, thus hindering the rotary compression process. For this reason, the present investigation involved determining the effect of tubular specimen wall thickness on temperature and its distribution. Fig. 4 shows the temperature distribution maps that were calculated by the FEM in the final stage of rotary compression for tubular specimens with various wall thicknesses. As predicted, the distributions of the temperatures are not uniform; the highest drops in the temperature can be observed in the region of the necking being formed. The non-uniform distribution pattern of the temperature in the region of the necking being formed results from the nature of the rotary compression process. During this process, the outside layers of the work piece are in contact with the colder tools, thus carrying heat away to the tools. In contrast, the inside layers of the work piece lose heat mainly due to heat exchange with the environment. The non- uniform temperature distribution in the region of the necking being formed results from that fact that heat is transferred at a more rapid rate to the tools than to the environment. The amount of heat transferred from the work piece to the tools and the environment is determined by the heat transfer coefficients, the magnitudes of which are given in the section describing the numerical model. Fig. 4. FEM-calculated distributions of the temperature in the final stage of the process for tubular specimens with various wall thicknesses The design of this forming process assumed that the external steps would not be in contact with the tools. As a result, they did not undergo deformation and their temperature was relatively uniform and high (above 1000 °C). Examining the distribution of the temperature in the zone where the tools have a direct impact on the tubular specimen, it can be observed that the highest drop in the temperature occurs when the tubular specimens with small wall thickness are used (low heat capacity of the tubular specimen). In addition, the forming of tubular specimens with relatively thin walls leads to the cooling of the material, practically over the entire section of the reduced step, and the temperature distributions in the reduction area are quasi-uniform. In contrast, the distributions of the temperature for tubular specimens with thicker walls differ: the temperature drops locally (on the surface). The farther it is from the surface layers that directly contact the tools, the higher the temperature. In effect, such a temperature distribution pattern enables the heat to flow from the inner to outer surfaces of the work piece, so that the temperature of the step is high enough for a relatively long time for the hot forming of this step. The rotary compression of tubular specimens with small thickness of their walls is, however, harder to perform, as the rapid and excessive cooling of the material across the entire section of the step being formed leads to a decrease in workability of the material and an increase in resistances to forming. 2 EXPERIMENTAL TESTS OF THE ROTARY COMPRESSION PROCESS FOR PRODUCING HOLLOW SHAFTS The numerical results were verified experimentally under laboratory conditions. To perform the experiments, we used a forging machine as shown in Fig. 5. With this machine, the process can be run in accordance with the pattern of the numerical simulations. Force and kinematics parameters of the rotary compression process were measured with a measuring system comprised of the following elements: a torque transducer to digitally record torque during forming; a displacement transducer to record velocity and position of the tools; pressure transducers to measure pressure in order to determine the force of the tools acting on the work piece. Fig. 5. Machine used in the experiments of the rotary compression process for hollow shafts The tubular specimens used in the experiments were C45 steel tubes of the same dimensions as those applied in the FEM simulation. The tubular specimens were first heated in an electrical chamber furnace to a forming temperature of approx. 1150 °C, then they were fed by tongs into a special feed mechanism to position the tubular specimen in the working space of the machine (formed by three rotating rolls - Fig. 6a). In the subsequent stage of the process, the tools travelled radially at a constant velocity v = 2 mm/s and were rotated in the same direction at a velocity n = 36 rotations per minute, consequently setting the tubular specimen into a rotational motion and, thus, reducing the diameter of the step in the centre of the work piece. Once the slides travelled the path of the required diameter reduction (^=D / d = 1.5), the translational motion was automatically stopped. After the actual forming operation, the work piece was sized into the desired shape between rotating rolls. At the final stage of the process, the tools were opened radially (Fig. 6b) and the finished part was removed from the working space of the machine in a special feed mechanism. The primary aim of the experiments was to verify whether the geometrical models applied in the numerical analysis were correct and to validate whether the rotary compression process is suitable for forming hollow stepped shafts. In the experiments of forming hollow tubular specimens with various wall thicknesses into shafts, we also analysed the geometrical parameters of finished products, process stability and forces. The experimental results obtained explicitly demonstrate that the proposed technique can be employed to form hollow tubular specimens into hollow stepped axes and shafts. The experimental results regarding the shapes and dimensions of produced parts agree well with both the theoretical assumptions (work piece shape assumed at design) and the numerical results. Fig. 7 shows the hollow shafts with various wall thicknesses produced in the experiments. As can be seen from Fig. 8, the thickness of the work piece walls increases, which is caused by the material's radial flow. Furthermore, it was observed that the length of all parts increased compared to the initial length of the tubular specimens, which results from the axial displacement of the material. It should however be stressed that the material flows axially mainly in the region where the tools have various radii, thus leading to a decrease in wall thickness in the region between the step being formed and the non-deformed steps of the work piece. The thinning of the wall can be significantly reduced or even eliminated by preventing free work piece elongation during the process. Fig. 6. Rotary compression process for producing hollow shafts; a) start of the process, and b) end of the process Fig. 7. Hollow shafts produced in the experiments from hollow tubular specimens with various wall thicknesses; a) general view, and b) axial sections In spite of the above, we can observe that the region between the end and central steps of the work piece is characterized by local thinning of the work piece wall. Apart from the local thinning of the work piece wall, a small arching of the axis of the shaft is noted, especially in regard to shafts formed from tubular specimens with thicker walls. This defect is caused by the lack of support for the work piece in the zone of the end steps that were not in contact with the tools during the process. Examining the geometry of the produced parts (Fig. 7), it can be observed that the initial thickness of a tubular specimen wall has a significant effect on the kinematics of material flow during the rotary compression process. The reduction in the outer diameter leads to a change in the wall thickness and work piece length. It is to be noted, however, that the relative increase in the wall thickness of the necking formed (compared to the initial thickness) decreases as the initial tubular specimen wall thickness increases (Fig. 8); what is more, in the case of the tubular specimen with the thickest wall (to = 11 mm), a small loss of 2 4 6 8 10 12 Initial thickness wall (0 [mm] Fig. 8. Variations in the wall thickness and work piece length as determined by FEM and in the experiments thickness of the wall in the formed step of the shaft can be observed. A reverse trend can be observed with regard to length of formed shafts. Here, the increase in the initial wall thickness leads to an increase in Time [s] Fig. 9. Torque in the rotary compression for hollow shafts with various initial wall thicknesses; a) experiment, and b) FEM Time [s] Fig. 10. Force in the rotary compression process for hollow parts with various initial wall thicknesses; a) experiment, and b) FEM length of the shaft. The observed relationship directly results from the differences in resistance to material flow that depend on tubular specimen wall thickness. In the forming of thin-walled tubular specimens, the resistances to the radial flow of the material are much lower than those observed when the material flow is axial, which leads to a rapid increase in the thickness of the walls of the shafts. With thicker walls, the material's axial flow is free from such resistances, which results in a rapid increase in the length of the shaft. The theoretical and experimental analyses were also performed to investigate forces and torques of the process, the variations of which are shown in Figs. 9 and 10. It can be noted that an increase in the thickness of a tubular specimen wall leads to an increase in the forces and torques. The nature of the variations in the force parameters is similar in all investigated cases. Initially, the forces and torques increase rapidly with the sinking of the tools into the material; in the steady-state stage of the process, the increase in the forces is stable and gentle. Then, at the final stage of the process, a rapid decrease in the forces and torques can be observed. There are, however, differences between the FEM-calculated variations in the forces and torques and those obtained in the experiments. The experimental results demonstrate that the forces increase much faster over time compared to the variations of the forces obtained in the FEM modelling. Moreover, the forces and torques are higher in the experiments (the differences can even amount to several percentage points in the case of tubular specimens with thin walls). The observed differences can result from the fact that the material cools faster in the experiments than in the numerical modelling. This is proved by the considerably lower differences in values of the forces obtained in the forming of tubular specimens with thicker walls. Owing to the higher heat capacity of the tubular specimens, the temperature of the tubular specimen is relatively high, thus ensuring sufficient material workability. 3 DISCUSSION AND CONCLUSIONS The majority of currently employed metal-forming techniques for hollow parts are highly complicated and relatively expensive; they only become cost-effective in mass production. The proposed new rotary compression method for hollow parts can be successfully employed in both small series and mass production, as it is based on the use of simple tools and devices. Furthermore, the use of hollow tubular specimens in rotary compression leads to lower consumption of both material and energy, which, in turn, results in lower production costs. The theoretical and experimental analyses of the rotary compression process have confirmed that stepped hollow shafts can be produced from hollow tubular specimens. The analyses led to a determination of the effect of work piece geometry on rotary compression and the geometry of finished parts. Furthermore, the force parameters of the process were investigated. Most importantly, however, the investigation helped validate the developed numerical model of the rotary compression process. Based on the results, the following conclusions have been drawn: 1. The numerical and experimental results show satisfactory agreement, which means that rotary compression processes can be modelled numerically. 2. Rotary compression can be employed to shape hollow tubular specimens of various wall thicknesses into stepped axes and shafts. 3. Parts produced by rotary compression processes exhibit higher wall thickness of steps and higher overall length of the work piece. 4. The rotary compression process for hollow parts involves non-uniform deformations; the greater the non-uniformity is, the higher the initial thickness of the tubular specimen wall is. 5. When tubular specimens with small initial wall thicknesses are formed into hollow parts, the material undergoes rapid cooling, which may hinder the rotary compression process. 6. An increase in the initial wall thickness of the tubular specimen leads to an increase in the tool force and torque. 7. Further research should be undertaken to determine failure modes and relationships between the technological parameters of the compression process and the geometry of tubular specimens and work piece. 4 ACKNOWLEDGEMENTS The study was conducted as part of the project "Qualifications for the labour market - employer friendly university", co-financed by European Union from European Social Fund. 5 REFERENCES [1] Opalić, M., Kranjčević, N., Habuš, S. (2011). Proof of strength of shafts and axles using finite element method. Transactions of FAMENA, vol. 35, no. 2, p. 63-71. [2] Neugebauer, R., Kolbe, M., Glass, R. (2001). New warm forming processes to produce hollow shafts. Journal of Materials Processing Technology, vol. 119, no. 1-3, p. 277282, D0l:10.1016/S09240136(01)00936. [3] Neugebauer, R., Glass, R., Hoffmann, M. (2005). Spin Extrusion - A New Partial Forming Technology based on 7 NC-Axes Machining. Annals of the CIRP, vol. 54, no 1, p. 241-244, DOI:10.1016/S0007-8506(07)60093-6. [4] Stanić, I., Galeta, T., Karakašić, M. (2008). Optimization of protecting wall tool for manufacture and assembly. Technical Gazette, vol. 15, no. 1, p. 45-49. [5] Pepelnjak, T. (2004). Numerical Analyses of Tube Hydroforming by High Internal Pressure. Strojniški vestnik - Journal of Mechanical Engineering, vol. 501, no. 1, p. 31-43. [6] Staupendahl, D., Becker, C., Weinrich, A., Hermes, M., Tekkaya, A.E. (2012). Innovative forming processes for tubes, profiles and sheets made of modern steel grades. Stahl und Eisen, vol. 132, no. 8 p. 47-54. [7] Jia, Z., Zhou, J., Ji, J.J., Yu, Y.-Y., Xiao, C. (2012). Influence of tool parameters on internal voids in cross wedge rolling of aluminum alloy parts. Transactions of Nonferrous Metals Society of China, vol. 22, suppl. 1, p. 21-26, DOI:10.1016/ S1003-6326(12)61678-1. [8] Dadras, P., Wells, W.R. (1984). Heat transfer aspects of nonisothermal axisymmetric upset forging. Journal of Engineering for Industry, vol. 106, no. 3, p. 187-195, D0l:10.1115/1.3185932. [9] Malinowski, Z., Lenard, J.G., Davies, M.E. (1994). A study of the heat-transfer coefficient as a function of temperature and pressure. Journal of Materials Processing Technology, vol. 41, no. 2, p. 125-142, D0l:10.1016/0924-0136(94)90057-4. Strojniški vestnik - Journal of Mechanical Engineering 61(2015)3, 157-166 © 2015 Journal of Mechanical Engineering. All rights reserved. D0I:10.5545/sv-jme.2014.2051 Original Scientific Paper Investigations on the Effects of Different Tool Edge Geometries in the Finite Element Simulation of Machining Lei Wan - Dazhong Wang* -Yayun Gao Shanghai University of Engineering Science, College of Mechanical Engineering, China This work focuses on the effects of cutting edge geometries on dead metal zone formation, as well as stress and temperature distributions in orthogonal cutting of P20 material using finite element method (FEM) simulation with sharp, chamfered, double chamfered and blunt tools. The cutting process is simulated with Arbitrary Lagrangian-Eulerian (ALE) approach in ABAQUS/Explicit. The simulation results suggest that the tool edge geometry influences the shape of dead metal zone considerably, while having little influence on the chip formation. An analysis of thermo-mechanical coupling was also conducted, and the results show that the stress distribution is affected by the temperature distribution and cutting speed because of the thermal softening effect and the strain rate hardening. A common analytical model is introduced to predict the residual stress, and equivalent Mises residual stresses are all calculated with four different tools to suggest that the tool edge geometry has a significant effect on the residual stress. The experiments are conducted using a CNC with former four kinds of tools at a speed of 480 m/min, and the residual stresses beneath the machined surface were measured with X-ray diffraction and electro-polishing techniques, and a chamfer tool at three different cutting speeds (250, 600 and 1000 m/min) to obtain the forces. The machining forces in both the cutting and thrust directions increases as the chamfer angle increases and decreases as the cutting speed increases. Keywords: tool edge geometry, coupled thermo-mechanical analysis, finite element method, dead metal zone, residual stress Highlights • Proposing a newly model of the tool edge geometry to analyze the effect of tool edge on the cutting process. • Four different kinds of tool edge geometries are consider in this analysis. • Finite element method (FEM) are utilized in cutting process. • Dead metal zone is formed under different tool edge geometries and varies with the shape of the tool edge geometry. • Stress and residual stress are influenced by the tool edge geometry. • The simulated results are validated with the analytical and experimental results and the forces obtained from simulation, analysis and experiments are compared. 0 INTRODUCTION Metal cutting is always considered to be one of the most complicated manufacturing processes for different materials. In order to attain sufficiently high production rates at minimal cost, optimization of cutting tool geometry is necessary. Moreover, the cutting process is greatly influenced by the cutting conditions, such as cutting velocity, chip thickness, and feed rate, as well as the tool geometry. The design of tool edge geometry influences process parameters, such as the shape of deformation zones, stresses on the chip and machined surface and the cutting forces mentioned in Fig. 1 [1]. These effects in turn affect the changes in chip flow and machined surface integrity. The modification of the tool edge geometry is referred to edge preparation. Fig. 2 illustrates four major types of edge preparation design used in most commercial cutting inserts: sharp edge; hone edge; T-land/chamfer edge and double chamfer edge. Moreover, Regions 1, 2, 3 are the missing regions of blunt, double chamfer and chamfer tools compared with the sharp tool, respectively. Fig. 2. Typical cutting edge preparation Fig. 1. Effect of tool geometry on performance parameters in turning Much research has been conducted about how the edge geometry of tools influenced the cutting process, analytically and experimentally [2]. Experimental results of a study done by El-Wardany et al. [3] show that for a sharp tool (compared with other geometries) the magnitude of the residual stress on the machined surface and the penetration depth of the stressed layer were reduced as cutting speed increased, while there is an opposite trend for the hone tools. Based on the large numbers of experiments, which were conducted with CBN tools for carburized hardened steel (600 to 720 HV) under both continuous and interrupted cutting conditions, Shintani et al. [4] and [5] analysed the effect of tool geometry on the cutting performance. They proposed that the optimum tool geometry for continuous cutting be specified as having a negative chamfer angle of 35° and a nose radius of 0.8 mm. Matsumoto et al. [6] investigated four different tool edge geometries (sharp, honed, single chamfered and double chamfered) on residual stress in precision hard turning, and concluded that tool edge geometry is the dominant factor determining the residual stress profile, and the residual stress with the honed and chamfer tool on the machined surface became more compressive. In recent decades, with the improvement of the computer technology, the finite element method (FEM) and numerical simulation have been increasingly used to study the machining process. Kountanya et al. [7] studied the effect of tool geometry and cutting conditions on experimental and simulated chip formation. Some observations can be obtained to illustrate that the machining forces increase as the tool edge radius increases, and the change of edge radius has little influence on chip morphology. Shatla et al. [8] applied the Lagrangian FEM to simulate the cutting process of H13 tool steel to investigate the influence of edge preparation (hone/chamfer) on tool temperature and stress. The simulation results showed that an increasing tool edge radius alters the distribution of temperature in the different kinds of tools, while the experiment results also indicated that the cutting force increases as the tool edge radius increases. When studying the influence of the tool geometry on the metal cutting process, one of the main points of interest is the existence of the dead metal zone under the chamfer or around the edge radius of the tool. The dead metal zone is trapped under the chamfer and almost entirely fills the chamfer, which acts as the effective cutting edge of the tool, thus protecting the tool surface from wear under various kinds of heavy cutting conditions. The drawbacks of cutting with a dead metal zone on the chamfered edge are that the forces on the tool increased, and the simulation results may be inaccurate due to the varying dead zone during cutting. Experimental and analytical studies discussed in [9] to [11] concluded that a dead metal zone is mostly dependent on the geometry of the chamfered part rather than the cutting conditions. Meanwhile, the dead zone is also formed with the blunt tool in the same way as for the chamfered tool, which is in good agreement with the results of assumption of the formation of BUE discussed by Waldorf et al. [12] for large-radius tools. The region of the missing zone determines the region of the dead zone. The surface finishing may depend greatly on the size and the shape of the dead zone. Al-Athel and Gadala [13] proposed a new volume of solid (VOS) method to simulate the metal cutting process with four different geometries tools. The simulation results suggest that the tool geometries have great influence on the formation of the dead metal zone, but little effect on the distribution of stress. Jacobson et al. [14] conclude that various kinds of dead zones near the tip of the tool may occassionaly become unstable in form and size, and sometimes crack partially or even entirely, resulting in deposits on the tool surface. From this background information, it can be determined that the above mentioned authors have the common view that the tool edge geometries significantly influence dead metal zone formation and residual stress distribution, but not the chip formation and the stress distribution. Aiming at these hypotheses, this paper presents a numerical analysis of a continuous chip formation process based on the finite element method for a better understanding of dead metal formation and chip formation. The effects of four common kinds of tool edge geometries on the formation of the dead metal zone and how they affect the distributions of stress and temperature have been studied. 1 NUMERICAL SIMULATION 1.1 Numerical Approach Numerical simulation of the cutting process can provide detailed results for process variables, such as stress, strain, strain rate, temperature, that are extremely difficult to measure with current technology. Owing to advanced technology and computer power, an alternative approach called the arbitrary Lagrangian-Eulerian (ALE) approach [13], which could combine the advantages of Lagrangian and Eulerian approaches, eliminating mesh distortion in Eulerian formulation and modeling the unconstrained flow of the chip in Lagrangian approach was applied in this research. The ALE approach is proved to be viable once a proper mesh motion scheme been implemented, Fig. 3 [15]. The simulation results are compared with the numerical work done by Movahhedy et al. [16], who simulated the cutting process via the ALE approach with four different tools whose rake angles were 0°. The ALE model (shown in Fig. 4) developed by Munoz-Sanchez et al. [17] allows the material to flow across an internal Eulerian zone surrounding the tool tip by using sliding Lagrangian and Eulerian contours. This method can avoid extreme mesh distortion and allows the simulation of a long machining time. The model is divided into four zones, allowing mesh motion or material flow across the fixed mesh. Zones 1 to 3 combine Lagrangian/ Eulerian boundaries with sliding boundaries where the material can flow tangential to the contour but not go across this boundary. Eulerian boundaries located at the entrance of Zone 1 and in Zone 2 can avoid the mesh distortion that is commonly revealed as the calculation advances. Zone 4 allows the material to flow across this region (which is a Eulerian region) with the mesh fixed. The main advantage of this technique is that it can avoid the extreme distortion in the region surrounding the tool tip. Therefore, it can be used to simulate a large machined surface in this paper. 1.2 Boundary Conditions and Material Properties Fig. 5 shows the schematic of the metal cutting process for a general case with a chamfer tool. The cutting force and thrust force are respectively defined by Fc and Ft . Fs and Fn represent the forces acting Fig. 3. Overall simulation approach along and normal to the shear plane, respectively. The main rake angle and chamfer angles are defined as a0 and aj . The length of the chamfered part is defined as bf, material flow (Vc) splits into two parts at the front of the cutting edge, one part shapes the chip at speed of Va, the other forms the machined surface at speed of Vb. Three deformation zones are shown in the graph. The metal cutting process is so complicated that the finite element model should be simplified, and some assumptions established as well. The assumptions of finite element model are: Fig. 4. Implementation of boundary condition and type of contour in the model; regions 1 to 3 combine sliding and Lagrangian/Eulerian boundaries; region 4 combines Eulerian boundaries 1. The workpiece material is isotropic. 2. The workpiece material accords with Von Mises yield criteria. 3. Machine tool bed and fixture are both rigid in the cutting process. which considers high temperature and high strain rate, usually presented with the equation below: ä = (A + B(š))(1 + Cln š )(1 -(T), (1) Fig. 5. Schematic view of metal cutting process with double chamfered tool with the T* is defined as: t *_ T - Tr T - T (2) where ст is the equivalent flow stress, e the equivalent plastic strain, e the equivalent plastic strain rate, e0 the reference strain rate, which equals 1 s1. The material characteristics are defined by the thermal softening coefficient m, the strain hardening exponent n, and constants adopted from [13], which are listed in Table 2. Tm and Tr are the material melting temperature (1480 °C) and reference ambient temperature (20 °C), respectively. In the actual metal turning process, cutting width is much larger than the back engagement of the cutting edge, so the chip deformation, which is perpendicular to the cross-section of tool edge, is approximately same along the tool edge direction, which is called the plane strain state. Therefore, the metal cutting process can be regarded as a two-dimensional orthogonal cutting model when simulating the cutting process by using the FEM. In this paper, basic geometry and boundary conditions of the numerical model are shown in Fig. 4, and a plain strain condition is assumed. The tool is fixed and cutting speed is applied to the workpiece to obtain the velocity field of the workpiece material, the rake angle is 10°, and the clearance angle is 7°. Material properties for the workpiece and tool are presented in Table 1. Table 1. Material properties for the workpiece and tool Material properties Workpiece Tool Material P20 Carbide tool Young's modulus [GPa] 210 530 Poisson's ratio 0.3 0.3 Conductivity [W m-1 °C-1] 51.5 120 Specific heat [J kg-1 °C-1] 470 343.3 Thermal expansion coefficient [°C-1] 1.4x10-6 5.2x10-6 In order to correctly simulate the cutting process with different tool geometry, it is necessary to introduce a material flow stress model to describe the material behaviour. The model is obtained from [18], Table 2. J-Cparameters for P20 A [MPa] B [MPa] C n m 178.5 462.4 0.0438 0.169 0.666 1.3 Frictional Model Whether the cutting simulation results are accurate and reasonable, to a large degree depends on the foundation of the frictional model; therefore, it is vitally important to choose a reasonable friction model. The rake face is divided into two workspaces, the sticking zone and the sliding zone, which is illustrated in Fig. 6. Mf Fig. 6. Curves representing normal and frictional stress distributions on the tool rake face Alvarez et al. [19] studied the effect of four different constitutive models and three friction coefficients (0.4, 0.6 and 0.8) on the simulation results. They conclude that the friction factor depends on the constitutive equations and their parameters, and its election is subject to the experiment results. Thus, when considering machining simulations of P20, a moderate friction coefficient (0.4) is used in this study and friction at the tool-chip interface is controlled by a Coulomb friction model which is expressed by the following relations: r = ^ —■—Analytical Simulation -a- Experiment ▼ Double Chamfer tool b) 0 10 20 30 Chamfer Angle [°] 40 300 250 [ 200 150 100 50 —■— Analytical - •- Simulation -At- Experiment ▼ Double Chemfer tool 0 10 20 30 Chamfer Angle [°] 40 Fig. 13. Comparison of a) cutting and b) thrust forces between analytical, experimental and simulation results of tools with different chamfer angles, and simulation results of double chamfer tool 0 0 0 225 ö 200 U U (3 h 175 '3 150 125 200 400 600 Cutting Speed [m/min] 800 1000 b) 200 , 175 150 125 200 400 600 Cutting Speed [m/min] 800 1000 Fig. 14. Comparison of a) cutting and b) thrust forces between analytical, experimental and simulation results of -10° chamfer tool and simulation with double chamfer tool under different cutting speeds that of increasing the hone radius. It can seen in Fig. 12 that the profiles of chamfer and double chamfer tool are almost same, but the peaks of them are both lower than the hone one. The peaks of the residual stresses obtained with the sharp tool are lowest among the other three tools, suggesting that the finer surface quality is acquired. Meanwhile, the experimental results from the operation with a -25° chamfer tool suggest that there is a little diference between measurment and prediction at the surface, which is partially due to the preexisting residual stresses on the actual machined surface that are neglected in the finite element model and the meshing quality in the simulations. 3.4 Effects of the Tool Geometry and Cutting Speed on Machining Forces Fig. 13 shows the effect of different tool geometries (sharp, chamfer, double chamfer) on cutting forces and the thrust forces obtained by simulations and experiments. It can be seen that the thrust force acquired with the -35° chamfer angle is much larger than other four forces, since the presence of a chamfer angle leads to the buildup of the dead metal zone where the workpiece material will be strained to a larger extent, compared with the sharp tool. Moreover, the force created by the double chamfer tool is between that of the tools with -10° and -25° chamfer angles, because the region of the dead zone created by the double chamfer is moderate. Without considering the beginning of the cutting process when the tool has not been loaded, the cutting force gradually advances and up to a relatively large point, then fluctuates in a small area; the thrust force has the same trend but changes more rapidly. The analytical results obtained from the Ren and Altintas [21] show a great difference between the experimental and simulation results, but give a clear description of the trends of forces which is in qualitative agreement with the others. Nalbant et al. [2] concluded that it is an increment-decrement relationship between cutting speed and min cutting force. Increasing cutting speed by 66.6 % (150 to 250 m/min) and 20 % causes the main cutting force to decrease by 14.6 and 10.4 %, respectively. This trend can be also found in Fig. 14, showing that the machining forces decrease as the cutting speed increases. Moreover, the cutting speed has more significant influence on the thrust force in contrast with the cutting force. Meanwhile, the tool edge geometry has little effect on the machining force as the cutting velocity reaches a comparatively high value. 4 CONCLUSIONS A finite element model for the simulation of chip formation with different cutting edge geometries during the cutting process has been presented. The simulation results show that the edge geometries affect the chip removal process less significantly, because the dead zones are formed under the chamfer, double chamfer and blunt edge to act as the main cutting edge of the tools, just like the sharp tool. The coupled thermal-mechanical analysis was conducted with four different tools, and some results were obtained. The behaviour and shape of the dead metal zone are studied by examining the distributions of the material velocity and the stress. The effect of tool edge geometry on the residual distribution was also studied, showing that the chamfer and double chamfer tool had the almost same function on the residual stress distribution in comparison with the hone tool and sharp tool. It is clear that the force in 165 the thrust direction has more dependence on the tool geometry and cutting speed, in comparison with the one in the cutting direction. The forces increase as the chamfer angle increases, and decrease as the cutting speed increases. The cutting edge geometry has little influence on the machining force when the cutting speed reaches a very high level. Finally, the present fulfillment of the ALE method is only applied to continuous chip formation, and can not directly extend to segmented chips, which need a better crack propagation scheme. A more complex mesh motion scheme should be used to allowed for the formation of new surfaces in chips. 5 ACKNOWLEDGEMENTS The authors would like to thank Shanghai University of Engineering Science (Project Code: 14KY0107) for providing financial support for the project, and the reviewers for their suggestions. 6 REFERENCES [1] Dogra, M., Sharma, V.S., Dureja, J. (2011). Effect of tool geometry variation on finish turning-A Review. Journal of Engineering Science and Technology Review, vol. 4, no. 1, p. 1-13. [2] Nalbant, M., Altin, A., Gökkaya, H.. (2007). The effect of cutting speed and cutting tool geometry on machinability properties of nickel-base Inconel 718 super alloys. Materials & Design, vol. 28, no. 4, p. 1334-1338, D0l:10.1016/j.matdes.2005.12.008. [3] El-Wardany, T.I., Kishawy, H.A., Elbestawi, M. A.. (2000). Surface integrity of die material in high speed hard machining, Part 1: Micrographical analysis. Journal of Manufacturing Science and Engineering, vol. 122, no. 4, p. 620-631, DOI:10.1115/1.1286367. [4] Shintani, K., Ueki, M., Fujimura, Y. (1989). Optimum cutting tool geometry when interrupted cutting carburized steel by CBN tool. International Journal of Machine Tools and Manufacture, vol. 29, no. 3, p. 415-423, D0l:10.1016/0890-6955(89)90010-2. [5] Shintani, K., Ueki, M., Fujimura, Y. (1989). Optimum tool geometry of CBN tool for continuous turning of carburized steel. International Journal of Machine Tools and Manufacture, vol. 29, no. 3, p. 403-413, D0l:10.1016/0890-6955(89)90009-6. [6] Matsumoto, Y., Hashimoto, F., Lahoti, G. (1999). Surface integrity generated by precision hard turning. CIRP Annals - Manufacturing Technology, vol. 48, no. 1, p. 59-62, D0l:10.1016/S0007-8506(07)63131-X. [7] Kountanya, R., Al-Zkeri, I., Altan, T. (2009). Effect of tool edge geometry and cutting conditions on experimental and simulated chip morphology in orthogonal hard turning of 100Cr6 steel. Journal of Materials Processing Technology, vol. 209, no. 11, p. 5068-5076, D0l:10.1016/j. jmatprotec.2009.02.011. [8] Shatla, M., Yen, Y.C., Altan, T. (2000). Tool-workpiece interface in orthogonal cutting-application of FEM modeling. Transactions of the North American Manufacturing Research Institution of SME, p. 173-178. [9] Kim, K.W., Sins, H.-C. (1996). Development of a thermo-viscoplastic cutting model using finite element method. International Journal of Machine Tools and Manufacture, vol. 36, no. 3, p. 379-397, D0l:10.1016/0890-6955(95)00054-2. [10] Shi, G., Deng, X., Shet, C. (2002). A finite element study of the effect of friction in orthogonal metal cutting. Finite Elements in Analysis and Design, vol. 38, no. 9, p. 863-883, D0I:10.1016/ S0168-874X(01)00110-X. [11] Movahhedy, M. R. (2000). ALE Simulation of Chip Formation in Orthogonal Metal Cutting Process, PhD thesis, University of British Columbia, Vancouver, p. 33-38. [12] Waldorf, D.J., DeVor, R.E., Kapoor, S. G. (1999). An evaluation of ploughing models for orthogonal machining. Journal of Manufacturing Science and Engineering, vol. 121, no. 4, p. 550-558, DOI:10.1115/1.2833050. [13] Al-Athel, K.S., Gadala, M.S. (2011). The use of volume of solid (VOS) approach in simulating metal cutting with chamfered and blunt tools. International Journal of Mechanical Sciences, vol. 53, no. 1, p. 23-30, D0l:10.1016/j.ijmecsci.2010.10.003. [14] Jacobson, S., Wallén, P. (1988). A new classification system for dead zones in metal cutting. International Journal of Machine Tools and Manufacture, vol. 28, no. 4, p. 529-538, D0l:10.1016/0890-6955(88)90065-X. [15] Liu, K., Melkote, S.N. (2007). Finite element analysis of the influence of tool edge radius on size effect in orthogonal micro-cutting process. International Journal of Mechanical Sciences, vol. 49, no. 5, p. 650-660, D0l:10.1016/j. ijmecsci.2006.09.012. [16] Movahhedy, M.R., Altintas, Y., Gadala, M.S., (2002). Numerical analysis of metal cutting with chamfered and blunt tools. Journal of Manufacturing Science and Engineering, vol. 124, no. 2, p. 178-188, D0l:10.1115/1.1445147. [17] Mu-oz-Sanchez, A., Canteli, J.A., Cantero, J.L., Miguélez, M.H.. (2011). Numerical analysis of the tool wear effect in the machining induced residual stresses. Simulation Modelling Practice and Theory, vol. 19, no. 2, p. 872-886, D0l:10.1016/j. simpat.2010.11.011. [18] Johnson, G.R., Cook, W.H. (1983). A Constitutive Model and Data for Metals Subjected to Large Strains, High Strain Rates and High Temperatures, p. 541-547. [19] Alvarez, R., Domingo, R., Sebastian, M.A. (2011). The formation of saw toothed chip in a titanium alloy: influence of constitutive models. Strojniški vestnik - Journal of Mechanical Engineering, vol. 57, no. 10, p. 739-749, D0l:10.5545/sv-jme.2011.106. [20] Liu, M., Takagi, J.-i., Tsukuda, A. (2004). Effect of tool nose radius and tool wear on residual stress distribution in hard turning of bearing steel. Journal of Materials Processing Technology, vol. 150, no. 3, p. 234-241, D0l:10.1016/j. jmatprotec.2004.02.038. [21] Ren, H., Altintas, Y., (2000). Mechanics of machining with chamfered tools. Journal of Manufacturing Science & Engineering, vol. 122, no. 4, p. 650, D0l:10.1115/1.1286368. Strojniški vestnik - Journal of Mechanical Engineering 61(2015)3, 167-175 © 2015 Journal of Mechanical Engineering. All rights reserved. D0l:10.5545/sv-jme.2014.2294 Original Scientific Paper Deterministic Mathematical Modelling of Platform Performance Degradation in Cyclic Operation Regimes Nenad Kapor1 - Momcilo Milinovic1- Olivera Jeremic1 - Dalibor Petrovic2 University of Belgrade, Faculty of Mechanical Engineering, Serbia 2University of Defense, Military Academy, Serbia This paper considers the modelling of extreme-capability working platforms that are operated in periodic cycles, each cycle having a predefined number of operations that affect the working surfaces. A novel hypothesis is introduced about the platform-degrading effects that cause an equivalent decrease in the successful operations after repeated cycles. Deterministic modelling, based on the basic equations of Lanchester and Dinner, is generalized here to include coupling between parameters. The newly developed mathematical model of performance degradation is in good agreement with both experimental measurements and numerical simulations. It is assumed that the new variables and their correlations link the Gaussian distribution and the observed performances of the testing platforms. Relative probability dispersions of affected surfaces are derived, as a new indirect referencing figure of merit, to describe simulations and compare them to experimental test data. The model proves a hypothesis that the degrading effects are a function of the platform capacity, frequency of operations and the number of available cycles. Degradation effects are taken into account through an equivalent decrease of effective operation capacities, reflected on the properties of the affected operating surfaces. The obtained estimations of degradation could be used in the planning of platform capacity as well as in the selection of real affected surfaces in various machining systems and for a wide range of different materials. Keywords: cycles, operations, extreme machine platforms, probabilities, deterministic modelling Highlights • Modelling of extreme-capability working platforms that are operated in periodic cycles. • Proposed methodology for performances degradation caused by operations composed in cycles. • Using Gaussian probability distribution law to predict degradation measures. • Predicting the changes of probability dispersion based on modelling and experimental data. • Determination of an analytical model based on a hypothesis that the degrading effects are a function of the platform capacity, frequency of operations and the number of available cycles. 0 INTRODUCTION Machines operating in cycles and their properties have not been studied in depth in literature and, as such, are not well described by integral mathematical models. If the effects of their operation are actions on the given working surfaces under given constraints, then the quality of the affected surfaces can be described by reliability functions. In this manner, the operating capabilities of the platform can be determined. A majority of the published papers use a standard approach to the measured performances that depend on the machine's designed purposes. Such processes are described in [1] to [3] for the abrasive flow machines (AFM) with which material is hardened by randomly treating the working surface with abrasive particles with polymeric fillers, and dispersed within the flow media. The authors of [1] classified the work piece parameters into three groups based, among others, on the number of cycles (operations) and the machining time. Some of these parameters were determined experimentally in [2], in which the authors recognized that the parameters denoted as the creeping time and the cycles frequency have impact on the quality of the machining process. In [3], the authors experimentally prove that the aforementioned parameters influence the process. Common for all three papers is that they do not include hidden random effects caused by particles affecting the surfaces in cyclic operations, although such effects significantly influence the quality of the surface treatment. In all three papers, there is no mathematical modelling of the process. Another similar type of machine with cyclic operation affecting working surfaces is described in [4] as shot-peening (SP) platforms. They bombard a surface with spherical beads to increase the material fatigue strength. The physical modelling of the influence of the bead shapes on the performance of the surface hardening process is presented in [5]. Random surface effects due to bombing cycles are a result of the quality of the machine's performance. However, the connection between the effects and the particular operations is missing in [5]. Paper [6] utilizes a risk function to consider the example of solar rays hitting a surface as a random process. In fact, the determination of the risk function dumping requires much more precise estimations of the distribution if probabilities of the effect occurrence on the attacked surface. However, this paper lacks a mathematical model of random disturbances of these probabilities. Common for the description of the processes in both AFM and SP machines, as well as the processes described in [6], is that they lack the deterministic or probabilistic mathematical modelling of cycles and their parameters on the efficiency of the final process. Mechanical engineering of extreme machines in defence technologies have particular operations grouped in the cycle regimes. These operations affect the working surfaces or areas, with constrained machine capacities with regard to operation numbers. The modelling of efficiencies in such cases is usually done using the deterministic differential theories of operational research. This approach is based on the so-called Lanchester and/or Dinner equations of particular probabilities and their distribution laws, as presented in detail in [7], as well as in [8]. Their equations use variable attrition rates as the frequencies in operations probabilities, similar to [9], in which surface point effects are taken with variable probabilities. The modelling of cycles' efficiencies in these references is performed with coupled equations, in which two subjects simultaneously affect each other. Their actions are interdependent but different; their efficiencies also evolve differently over time. This approach is not entirely useful to define a standalone efficiency estimation for a single subject. A mathematical model of the equipment with constrained capacities that generates identical repeated operations in a given order is presented in [10] for air platform equipment. The main contribution of this paper is the treatment of the action on the working surfaces as a random process, but the probability distribution laws on the affected surfaces are missing. The two-dimensional Gaussian distribution laws, used for welding processes, as referred in [11], could be useful in the estimation of random processes on surfaces. According to the state-of-the-art analysis as presented in the cited papers, there is no comprehensive mathematical model, proven by experimental data, that would be capable of explaining relations between the machine's cyclic performances, its capacities and the quality of randomly affected surfaces, as well as the designed processing time. This is because the papers do not consider two repeating processes simultaneously acting on a single object (one as working and the second as redundant or parasitic), which together change the quality of the expected performances. The objective of this paper was to develop a general joint mathematical model that includes all pertinent factor that influence the final efficiency of processing, thus enabling the simulation and evaluation of these parallel processes. Based on the specific requirements for sequential processing of the surfaces, a mathematical model is developed using a deterministic approach treating the surface processes as random variables. The objective was to test the efficiency of cyclic operations affecting the working surfaces, essentially by considering differences caused by the capacities and operation rates of the processing machine. This was shown using the experimental data on operation platforms with extreme performances. The novelty of the approach presented in this manuscript is the redesign of coupled deterministic equations done in a new manner. In the literature above, these equations are employed to describe the mutual effects of objects as a function of the elapsed process time. This approach in the literature makes the time functions dependent on the performances of two objects. In our approach, one object executes two operations in parallel, one of them comprising the working process itself, and the other, parasitic, occurring as self-degradation dependent on the first one. Both happen on the same object, i.e. the operation platform performing the same action. The new approach composes deterministic equations to describe this and to measure changes in the platform efficiency. With our approach, the quality of the working process is the convolution of both kinds of operations in one cycle, as well as their frequencies. The number of cycles influences the random arguments and reflects the probabilities of working surface coverage that follow the two-dimensional Gaussian distribution laws. This was taken as the measure of the changes in quality due to self-degradation. 1 EXPLANATION OF THE GENERAL MODEL The model offers the possibility of evaluating the degradation of the platform performance, with regards to the equipment and devices contained within the platform. The quality of the affected surface is regarded as the dimension of probability dispersion. This dimension appears during the execution as the consequence of cycle duration and the operations frequency, as well as of the capacities of the platform. The approach presented in [12], which developed operations frequency coupled with execution probabilities as the combined attrition parameter, was used in developing our general model. Changes of probability dispersions of random values on the affected surface appear in the form of the Gaussian distribution law. The degradation of the platform properties through the operation cycles is represented by changes in the Gaussian distribution. This is valid under the assumption that one particular Gaussian distribution function is valid for each cycle in the working regime. In our approach, this function is distributed in successive cycles in the form of extended probability dispersions of both random arguments in the two surface directions. Consequently, the changing of efficiencies over time is measured by the resulting effects on the new randomly affected surfaces. The decrease in the efficiency with each new cycle in reflected in the new less-affected surfaces. This also results in the degradation of the working platforms' capabilities caused by less effective particular operations in the cycles. The cause of this degradation could be a consequence of rapid high-energy operations realized in short-time sequences (high mechanical power values) in successive, orderly repeated cycles, similar to those described in paper [13]. However, according to that paper, the affected points on the surfaces do not obey any probabilistic law, and thus there is no error distribution as a modelling parameter. In our research, we use the changes of the probability dispersion (PD) after each cycle, due to all the errors in the cycle, as a measure of the platform efficiency. These changes are caused by the generator of the cycles and by its self-degradation, and are reflected in the decreased number of declared operations. This makes the designed operational capacities of the platforms less effective with the number of cycles. In order to estimate the degraded platform performances by means of time-based simulation, new relative parameters have been accepted in the modelling. The deterministic modelling of the estimations of the so-called vulnerability performances is presented in [14] and [15]. The performances considered there are similar to our degrading platforms' performances. The models presented in [16], called Pexpot, Levpot and Dynpot, were also developed as vulnerability considerations based on the attrition rate function and thus indirectly describe the kind of expected degradation capabilities. An essential difference of our model is that the degradation of the system appears as a direct consequence of self-degradation caused by the effects of the repeated cycles. The designed frequencies and functional probabilities, contained in each operation, are reflected in the full platform capacity on the affected surfaces. This effect makes the proposed model more useful in planning the redesigning of platform capacities for required affected surfaces. 2 MATHEMATICAL MODELS In the presented model, the platform has the capacity of Mptj particular operations oriented toward the working surface. These operations occur in dynamic regimes with successive frequencies X and a probability of surface action of approximately p = 0.997. This is provided using the maximum technical dimensions of the surface, which correspond to the 64 PDav2. The width and depth of the surface used eight of the same average probability dispersions PDav, in both surface directions. Average probability dispersions PDav are taken as an equal of the expected Gaussian distribution of two-dimensional random arguments. The probability variations are represented as functions of the cycle number and of the full capacity of operations. The designed properties of these processes are consequently the function of probability changes. The adopted hypothesis is that the degradation of the platform performances is an imaginary effect, able to be explained by the values of the effective and ineffective numbers of operations. This ensures a possibility of considering the ineffective number to be a value increasing with the number of cycles during the exploitation time. In that sense, the increasing number of ineffective operations corresponds to the increase of cycle probabilities dispersions. Operative consumption is realized in cycles with the same sequential probabilities of operations, p, as in [10]. In that case, the frequency of executions of real operations, as the real rate of operation, is: dM dt p = -Xp = -a p. (1) This determines the remaining number of operations as the Mp=Mp (t) in each moment of time t in the cycle duration interval. It is expected that the probability would not have a fixed value but would vary over exploitation time. The changes of probability function p could mean random changeable performances that disturb the rate of real operations on the working surfaces. The changes in probabilities affect the rate of real operations Mp in Eq. (1). This is not really possible because the frequency of operation executions is a designed property of the platform hardware. The present hypothesis has only an imaginary effect. The acceptable solution could be to recalculate the influence of the number of ineffective operations on the new probable dispersion PD reflected in a new Gaussian distribution but for the unchanged execution operations probabilities. The consequence is that the model has to consider the extended working surfaces, with new dimensions 64 PD2 engaged in operations after each cycle. The platform degradation, as an imaginary effect, is a process in the real cycle time and is simultaneously parasitic in real operations. A new value of the modified equivalent number of operations mp(t) is diminished by this imaginary effect. This is generated as a current and recalculated capability of working platform. The new value is lower than the real number of the remaining operations Mp(t). At the very beginning, it is equal to the real available capacity mp0 = Mpo. The reason is underpinned by the fact that the model of self-degradation is viewed as a new, fictive rate of equivalent non-effective operations changes mp, which is not equal to the rate of real operation Mp . This orients the mathematical model to consider the share of degraded value on each of the real operations and, by that effect, redesign the remaining number of operations available on the platform. Such a transformation implies that the degrading rate of mp and the real rate Mp during each cycle are proportional to the remaining equivalent dimensionless number of operations 1/mp. The correction coefficient is the portion of one operation within the actual remaining equivalent number mp. Based on the previous concept, the differential equation for the degradation rates, Eq. (1), becomes: dmr 1 dt ■ = -ar (2) Since the model of equivalent numbers is a function of time and the current equivalent number mp , as the instantaneous remaining capacity, the platform performances are degraded continually with each cycle. This always means a new valid number of operations with regard to the remaining capacity. It is inappropriate to use the approach as constant and fixed for any platform capacity since it is dependent on the available number of operations. The relative degradation of the platform capacity is taken as more acceptable in modelling with the functional ratio pp(t)=mp(t) /m where цp (t = 0) = 1, as the current relative capacity degradation of the platform. The general differential equation (Eq. (2)) of new functional Цp, by methodology given in [7], is: dßp dt ■ = -a p VpMpp0 (3) If the platform, under the same conditions, executes repeating working cycles n times, in equal time intervals for each cycle of At , and the rate of operations in the cycle is the same, then using Eq. (3), any of i cycles (where i = 1, 2, ..., n) is used at the beginning of a new, redesigned equivalent number of operations from the previous cycle. The current relative degrading in cycle is defined as a new differential equation: d Hp. 1 Pi 1 ■ — (4) - = -ap dt p i = 1,2,3. HP mp pi p(i-i) The solution of Eq. (4) is: Hp.(t ) = 1 — 2a r (5) p The function of the current relative capacity degradation of the platform full capacity after (i-1), and during the ith cycle at an instant (i-1)At < t < i At, similarly to [10], is: i-1 Hp (t) = Hpi (t)U^P( у (6) j=1 (j) with the condition j=i p( j) for the i = 1. The estimation of the relative efficiency of the current process is the function of the affected and the initial working surface. This functional is determined for the unaffected, remaining surface at each cycle and the final working surface from the previous cycle, taken as initial in the current one. It is given in the form: H p. (t) = 1 - M). S(i-1) (7) The differential equation of the relative efficiency of the current process, as the remained relative surface within operation cycle considered as the degraded ones, according to a similar differential equation in [7] and [10], is: dHc, dt - = -Ui »Pi Hc; : 0 < t < At. (8) If the platform operates in cycles without degradation effect, its /лр = 1= the functional /лр t. m does not affect Eq. (8) and the coupling of Eqs. (4) and (8) is lost. Consequently, the relative efficiency of the current process, denoted by n* , during the un-degrading surface processing in the cycles is described by: TT * dt (9) In both Eqs. (8) and (9), the operation number in one cycle is the designed capability, and could be variable. This depends on the designed cartridge capacity used for continual operations in the short impulse regimes. The well-balanced example between the number of operations and the covered affected surface in one cycle is the referent platform given in [10]. It uses cartridges of maximum Np(At) = 8, and its cycle expires in 4.4 seconds. The accepted functional designed capability of the platform, redesigned for the considered example, is: U i =Ър 1.82a (10) Appropriate values need to be calculated for each platform cartridge with their declared performances regarding the expected affected surface. The solution of Eq. (8) is: „2 U\m PH r 3 - ^ (t W. (0)e Pi (11) while for the ideal, un-degraded effect, from Eq (9), it takes the form: nI () = n* t0К (12) The next appropriate assumptions for the initial conditions are used: • Model A with initial conditions for the relative efficiency of the current process at the very beginning i (0) = 1 Si (°) S(i-1) = 1. Model B with variable relative efficiency of the current process at the very beginning Hc. (t = 0 ) = nc. < 1 taken in the next cycle from the end of the previous one. In both cases, the designed cycle capability of the platform , 1.82a , . . „ Uimax = A p-= const and is constant in all So cycles as a declared value. The cumulative relative efficiency of all finished processes is: Vc ;=1 (i) (13) For the experimental verification of the correlation between the platform capacities and the affected surfaces for the degraded as well as for the un-degraded (available) number of operations, new expressions were required. If all available cycles on the platform are expired, the full affected surface Sp and the working surface S0 can be correlated. The correlation could be expressed via a relation analogous to Eq. (7), using degradation effects on the surface given by cumulative relative efficiency of all finished processes in Eq. (13). This yields the relation So = . (14) 1 -Vc The same logic, analogous to Eq. (14), could be used for the non-degraded working surface S0 and the affected surface Sp equations. They also have to be related to the cumulative relative efficiency of all finished processes without degradation of n*, whicn is a product similar to Eq. (13), with n*© determined from Eq. (12), in the same form as Eq. (14). For the ideal designed number of operations, as well as for the degraded number, the particular affected surfaces are equal ( S* = Sp). This is because no operation in the cycle is missed but is only exposed somewhere on the larger area. The working surfaces S0 and S*0 could be determined as the squares of the appropriate Gaussian linear values of average probability dispersion PDav for both cases. Since the degraded and the non-degraded cumulative relative efficiency of all finished processes are the functionally correlated to the surfaces given by Eq. (14), the dispersion values also satisfy these correlations in both cases. Since the surfaces are taken as 8PDav x 8PDav, using basic Eq. (14) and their analogues, the ratio of PD* in un-degraded and PD of degraded cases is expressed by: PD * PD 1 -Hc 1 (15) This new approach provides the method for a comparison of degraded and un-degraded surfaces of accepted dimensions, or for treating expected degrading by constraining the allowable ratio. o 3 SIMULATION DATA AND RESULTS Simulation tests are realized using MATLAB software package and compared with experimental research. The basic assumptions in the simulations were: • Platform capacities and cycles performances presented in Table 1 are used for simulation testing, as well as in experimental modelling. These data are used in the platform degradation modelling; • Numerical test is provided for the 8, 16, 24, 32, and 40 operation capacity of the platform. As it was accepted in the mathematical model, one cycle had 8 operations and expired in a nominal time of 4.4 seconds. Fig. 1 represents the simulation results of the current relative degradation of successive ordering operations in the sequential cycles. The degraded values are positive and decreasing, bounded with value 1 from the upper side. The simulation shows that at the end of each cycle, the current relative degradation final value is decreasing, regardless of the number of operations remaining constant. The consequence of the model is that platform capability degrades more rapidly in each subsequent cycle, regardless of the same number of operations. It is the consequence of the decreasing number of equivalent operations after each real execution rate. The rapidest decrease of the current relative capacity degrading has been observed for the platform with 8 designed operations within one cycle. Lower degradation occurs on platforms that have 16, 24, 32 and 40 operations as the full capacity. Platforms with higher initial capacities, e.g. the number of operations, have smaller gradients and lower degradation at the end of the cycle. Fig. 2 represents the equivalent number of degraded operations starting with a different real number of platform capacities. It is visible that the gradient disperses with time. This process shows the minimum gradient for the platforms tested with 40 or more real operations compared to platforms with fewer capacities used in the simulation model. W/lvvlv/f t* Vi/ ! v у < -«-8 Mp -•-16 Mp 32 Mp -40 Mp 1 0 10 20 30 40 50 60 TO 80 90 100 t И Fig. 1. Functions of current relative degradation of successive ordering operations in the sequential cycles t [s] Fig. 2. Real initial and degraded equivalent number of operations on the platforms with different operation capacities Fig. 3 explains the model of cumulative relative efficiency of all finished processes. As it shown in Figs. 3a and b, the unaffected surface decreases with the increasing number of operations. This decrease is more or less similar to the experimental results of surface roughness decrease, presented experimentally in [1]. This paper has shown similar surface effects as our model. Table 1. Simulated platforms performances Capacity variant Number of operations in one tool set The number of tool-sets Available number of platforms operations Rate of operation X [1/s] Working surface [m2] Effective tool radii Rc [m] Elementary surface efficiency a [m2] Number of designed affected operations in one cycle 1 4 2 8 1.8 1.7 0.126 0.05 8 2 4 4 16 1.8 1.7 0.126 0.05 8 3 4 6 24 1.8 1.7 0.126 0.05 8 4 4 8 32 1.8 1.7 0.126 0.05 8 5 5 8 40 1.8 1.7 0.126 0.05 8 Paper [1] also reported significant changes in the first 20 operations (so-called cycles in [1] and [2]) and the gradual reaching of a saturated level. The result shown in Fig. 1 also shows the saturation performance of relative degrading, but with at least 40 operations. It could be seen that those values are of the same order of magnitude. Experiments in [1], developed for surface roughness, have shown that the roughness decreased slightly after 50 operating cycles. Comparative considerations about similarities of surfaces affecting processes, shown in Fig. 3, regardless of whether they have been tested on the different machines with different purposes, showed the same sort of behaviour. Two models of surfaces with different initial conditions marked as a) and b) shown in Fig. 3 are tested in simulation. The first tested case, for the both models, is degraded platform performance and the 4 INITIAL CONDITIONS ■A model degrade 40 operations capacity ■A model un-degraded 40 operations capacity A model degrade 32 operations capacity ■A model degrade 24 operations capacity —A model degrade 16 operations capacity —A model degrade 8 operations capacity V-* -Ä^s [o] 1 0.9 0.8 0.7 0.6 „ 0.5 NITIAL CONDITIONS B model degrade 40 operation capacity B model un-degraded 40 operation capacity B model degrade 32 operation capacity -■■X-- B model un-degraded 32 operation capacity B model degrade 24 operation capacity B model un-degraded 24 operation capacity B model degrade 16 operation capacity B model un-degraded 16 operation capacity B model degrade 8 operation capacity >kX --Ф--1 I a) [s] b) Fig. 3. Current relative efficiency of affected surface with variable operation designed capacities and number of cycles; a) model A, and b) model B [o] 0.9 0.8 0.7 0.6 0.4 0.3 0.5 0.4 0.3 0 5 10 10 15 Table 2a. Experimental data for the probability dispersions Cycle No Model B: Experimental test of PDav data [m] Model B: Dispersion error [%] Model A: Experimental of PDav data [m] Model A: Dispersion error [%] 1 7.5 36 5.0 32 2 5.0 10 5.0 32 3 5.0 10 9.0 22 4 5.0 10 9.0 22 5 5.0 10 9.0 22 6 5.5 0 7.5 1 PDav 5.5 12.67 7.4 21.8 Table 2b. Simulation of the probability dispersions Total cycles time t [s] Degraded Цо [°] Non-degraded Цо* [°] 1 - Цо [°] 1 - Цо* [°] PDav* / PDav % of surface extension Operation capacity Цо Ф 1 Model B 4.47 0.936 0.769 0.064 0.231 0.526361 47.36 8 8.94 0.468141 0.394497 0.531859 0.605503 0.937217 8.675 16 13.41 0.117549 0.100695936 0.882451 0.899304 0.990586 2.84 24 17.88 0.012616 0.010803896 0.987384 0.989196 0.999084 1.217 32 22.35 0.000476 0.000401542 0.999524 0.999598 0.999963 0.608 40 Цо = 1 Model A 4.47 0.936 0.769 0.064 0.231 0.526361 47.36 8 8.94 0.572717 0.482932 0.427283 0.517068 0.909042 10.67 16 13.41 0.288294 0.247261184 0.711706 0.752739 0.972362 4.36 24 17.88 0.120871 0.103355175 0.879129 0.896645 0.990184 2.4 32 22.35 0.041622 0.035244115 0.958378 0.964756 0.996689 1.556 40 second one is the non-degraded performance. Both curves of the cases, for both models, are shown in Figs. 3a and b. The platforms with lower operation capacities, 8 and 16, in simulations have greater differences for degraded and un-degraded tests for Model A. This is also valid in Model B. At higher operation capacities, 24, 32 and 40 have diminishing differences except in Model B where their differences disappear more rapidly than in Model A. The curves from Fig. 3 present the relative efficiency of the current process, with regard to cycle time as a continuously varying value. Differences between the relative values of affected surfaces in the degraded and un-degraded cases have been recalculated on their new extended working surface dimensions and transformed into their probability dispersions ratios. The average probability dispersion, PDav, as the measure of degraded surface is shown in experimental and simulation test data in Tables 2a and b from Eq. (15) for the platforms with different capacities and cycle numbers composed in the tool sets (Table 1). Repeated experimental tests have been realized in cycles with 8 operations in 6 realized experiments shown in Table 2a. It is obvious that both experimental models, A and B, show strong degradation of the initial probability dispersions. The decrease is slowed down by the increasing number of operations in the new cycles. Saturation is achieved after 24 operations for Model A and after 16 operations for Model B. It is obvious that Model B, where the relative efficiency of the current process at the very beginning of cycles is taken from the previous cycle, has a smaller saturation decrease of PDav in both experimental and simulation cases. In Table 2b, the dispersion in Model B is saturated after 24 operations from 47 to 2.84% and further approaches 1%. In both models, after at least 40 operations, (in experiments after 48), the probability dispersion PD stopped decreasing and diminished. The percentage of the errors measured by the surface extension over PD, for the model in simulations and in experiments, is about 4% for Model B and about 29% for Model A. This is because the decreased unaffected surface from the previous cycle in Model B is repeatedly taken as an initial surface for each next cycle. Consequently, errors increase as the cumulative values from the previous cycles. Taking into account the unsteady-state random processes on the designed platforms, predictions of efficiency made by this indirect modelling are of sufficiently quality. 4 CONCLUSIONS In the present state of the art, in the field of surface-affecting machines, there is no unified model theory that connects the variable machine performances with effects on the affected surface determined in continuous working process. Operational research modelling offers some models of interaction of two interdependent objects. In our model, one object, a surface affecting machine as a platform operating in cycles, is self-degrading due to the effects of sequentially repeated cycles. In our hypothesis, it was assumed that the rate of degradation is variable and proportional to the rate of real operations divided by the actual remaining capacity. This novel approach is proved indirectly for the effects occurring on the working surface. These effects are measured by determining the statistical variations of the surface parameters. The relevance of this model is that the real effects by the cyclic operations and also by the planned operation capacities could be predicted for the required surface effects. The model is also able to predict the extreme machine performances based on the expected quality for the surface treatment. The presented simulation results are consistent with the experimental data and coincide with the research cited. Further research in this area could involve an extension of the experimental data base in order to improve the simulation model according to the type of special extreme machines and their affected surfaces. The model can be readily applied to additional processes and effects. This could be done by exponential redesign of the basic correlations between the rates of operations and the rates of degradations, or by including variable frequencies and probabilities. 5 ACKNOWLEDGMENTS This paper is a part of the research within the Project III 47029 for MNES-RS in 2014-2015. 6 NOMENCLATURE Mp(t) current number of operations, mp(t) current number of degraded operations, Mp operation consumptions rate, hp rate the fictive changes of non-effective operations(rate of equivalent operations), M p0 capacity of particular operations, Po initial degraded number of operations, mp(i-l) initial degraded number from the previous cycle, N number of cyclic operations, mp (t) current number of operations during So sp S (i-i) degradation, remaining non-degraded surface, affected surface remaining non-degraded surface after the (i-i)th cycle, S(t) current non-degraded surface, At cycle period, PD* probability dispersion of un-degraded, operations on the platform, PD probability dispersion of working surface in degraded platforms operations, Ui designed capability of the platform, ap = Xp attrition rate of operations (consumption rate of operation numbers), p probability of each operation, Hp. (t ) current relative degradation in the ith cycle Hc. (t) relative efficiency of the current process, X=-Np(At) / t operation frequency for a given cycle, Np(t) number of operations per cycle, a elementary efficiency on the surface, ßp(t) current relative capacity degradation, Гс relative effic iency of the current process without degradation, jjc (relative efficiency of the current process, after i-i cycles with degradation effect, cumulative relative efficiency of all finished processes. 7 REFERENCES [1] Gov, K., Eyercioglu, O., Cakir V.M. (2013). Hardness effects on abrasive flow machining. Strojniški vestnik - Journal of Mechanical Engineering, vol. 59, no. 10, p. 626-631, D0l:10.5545/sv-jme.2013.1129. [2] Kar, K.K., Ravikumar, N.L., Tailor, P.B., Ramkumar, J., Sathiyamoorthy, D. (2009). Performance evaluation and rheological characterization of newly developed butyl rubber based media for abrasive flow machining process. Journal of materials Processing Technology, vol. 209, no. 4, p. 22122221, DOI:10.1016/j.jmatprotec.2008.05.012. [3] Rhoades, L. (1991). Abrasive flow machining: a case study. Journal of Materials Processing Technology, vol. 28, no. 1-2, p. 107-116, DOI:10.1016/0924-0136(91)90210-6. [4] Zupanc, U., Grum, J. (2011). Surface integrity of shot peened aluminium alloy 7075-T651. Strojniški vestnik - Journal of Mechanical Engineering, vol. 57, no. 5, p. 379-384, DOI:10.5545/sv-jme.2010.142. [5] Fang, L., Zhao, J., Li, B., Sun, K. (2009). Movement patterns of ellipsoidal particle in abrasive flow machining. Journal of Materials Processing Technology, vol. 209, no. 20, p. 60486056, DOI:10.1016/j.jmatprotec.2009.08.013. [6] Lugarić, L., Majdandžić, Lj., Škrlec, D. (2010). Countrywide positioning of domestic solar water heating systems using risk analysis and geographical information system. Strojniški vesnik - Journal of Mechanical Engineering, vol. 56, no. 1, p. 3-17. [7] Przemieniecki, J.S. (2000). Mathematical Methods in Defense Analyses, 3rd edition. American Institute of Aeronautics and Astronautics, Redstone. [8] Maybee, J.S. (1985). The theory of combined-arms lanchester-type models of warfare, Naval Research Logistic Quarterly, vol. 32, no. 2, p. 225-237, DOI:10.1002/nav.3800320204. [9] Anderson, L.B. (1993). A heterogeneous shoot-look-shoot attrition process. Simulation and Gaming, vol. 24, no. 3, p. 277-293, DOI:10.1177/1046878193243001. [10] Petrovic, D., Milinovic, M., Jeremic, O., Kovac, M., Kapor, N. (2014). Air force support operations and comparative performances. Symposium of Operational Research, p. 16-19. (in Serbian) [11] Ghosh, A., Barman, N., Chattopadhyay, H., Hloch, S. (2013). A study of thermal behaviour during submerged arc welding. Strojniški vestnik - Journal of Mechanical Engineering, vol. 59, no. 5, p. 333-338, DOI:10.5545/sv-jme.2012.775. [12] Scales, J.R. (1995). A modified lanchester linear process calibrated to historical data. Warfare Modelling, 3rd edition, Military Operations Research Society, Alexandria, p. 345-356. [13] Milinovic, M., Jeremic, O., Kovac, M., Petrovic, D. (2014). Discrete deterministic modelling of autonomous missiles salvos. Defence Science Journal, vol. 64, no. 5, p. 471-476, DOI:10.14429/dsj.64.5791. [14] Zipfel, P.H. (1995). Modelling and Simulation of Aerospace Vehicle Dynamics. American Institute of Aeronautics and Astronautics, Reston. [15] Ball, R.E. (1985). The Fundamentals of Aircraft Combat Survivability: Analyses and Design. American Institute of Aeronautics and Astronautics, Redstone. [16] Anderson, L.B., Miercort, F.A. (1995). On weapons scores and force strengths. Warfare Modelling, 3rd edition, Military Operations Research Society, Alexandria. [17] Law, A.M., Kelton, W.D. (1982). Simulation Modelling and Analysis. McGraw-Hill, New York. Strojniški vestnik - Journal of Mechanical Engineering 61(2015)3, 176-186 © 2015 Journal of Mechanical Engineering. All rights reserved. D0l:10.5545/sv-jme.2014.1914 Original Scientific Paper Selection of Optimum Process Parameters in High Speed CNC End-Milling of Composite Materials Using Meta Heuristic Techniques - a Comparative Study Vikas Pare - Geeta Agnihotri - Chimata Krishna* Maulana Azad National Institute of Technology, Bhopal, India The machining of aluminium metal matrix composites in CNC high speed conditions is significant because such composites have diverse applications in the aeronautics industry. Because that industry requires high quality outcomes, the prediction of surface roughness, which depends on input process parameters, assumes significance in the maintaining quality of products. Even though many researchers have worked in the area of conventional machining, very few of them have explored optimization techniques, such as teaching-learning-based optimization (TLBO) and gravitational search algorithms (GSA) in high speed environments. In this research, an attempt is made to determine the optimum machining conditions for the end-milling of composite materials using GSA. Input process parameters, such as cutting speed, feed, the depth of cut and the step-over ratio are taken as independent variables, and surface roughness is taken as dependent variable. Experiments were conducted on Al2O3 + SiC metal matrix composite by considering selected variations in the input process parameters. Surface roughness is measured in each of cases, and the required data is obtained for further analysis. An empirical relationship is established between dependent and independent variables in the form of linear and non-linear regression equations, and the results are analysed. The results showed that GSA gives better results for surface roughness when compared to the genetic algorithm, simulated annealing and TLBO methods. An additional set of experiments was conducted to validate the results obtained. Keywords: high-speed machining, metal matrix composite, surface roughness, gravitational search algorithms, teaching-learning-based optimization Highlights • Preparation of composite material of best mechanical properties. • Microstructure analysis to study the machinability of composite material. • Modelling of input parameters with surface roughness for high speed CNC machining of composite material. • Optimization of surface roughness using gravitational search algorithm. • Validation of results by conducting additional set of experiments. 0 INTRODUCTION High speed milling (HSM) has assumed importance in recent years due to increased demand for quality, productivity and cost reduction in manufacturing [1]. HSM can be used mostly for relatively softer materials and the production of components in mass scale. The machining of the aluminium metal matrix composite is a significant high-speed milling application. This technology has wide application in the aeronautics or aerospace sectors, and the moulds and die industry [2] and [3]. A metal matrix composite is made by combining at least two constituent parts, one of which must be a metal. Normally, other constituents may be different materials, such as ceramics or organic compounds. In this paper, silicon carbide is used as second constituent, as it reduces the cost of resultant composites. In order to obtain the required properties in the matrix alloy, the reinforcement material, the volume of the reinforcement, location and the shape of the reinforcement and fabrication method can all be varied [4]. The objective involved designing a metal matrix composite material mainly by adding the desirable attributes of metals and ceramics. The steps involved with castings of metal matrix composite are as follows: • Melting the aluminium metal with 5, 10, 15 and 20 % on a mass fraction basis; • Pouring it into an already prepared mould or cavity, which has the shape of the desired component; • Cooling and solidifying the molten metal in the mould; and • Removing the solidified component from the mould and cleaning it properly. The machinability of aluminium composites is considerably high in comparison to other materials. Whether the material is stable after a high-speed milling operation is performed on it needs to be studied; i.e. the structure of material needs to be checked after machining. Surface roughness is a measure of the technological quality of a product and has influence on manufacturing cost and the quality of the product. Therefore, industries always choose to maintain the good quality of the machined surface. The surface roughness and material removal rate considerably change with the variation of cutting process parameters; thus, the appropriate selection of process parameters has significant role in the prediction of the surface finish and material removal rate in the high-speed end-milling process. Therefore, it needs to be studied in the context of the machining of aluminium composites as well in high-speed machining conditions. Theoretical models have been proposed for the selection of process parameters by earlier researchers [5] to [7]. However, they do not yield good results under all the conditions of experimentation. As a result, machine operators generally use trial-and-error approaches to set up milling machine cutting conditions in order to achieve the required surface roughness. These methods are not productive and are extremely time consuming. Therefore, establishing empirical relations between output parameters, such as surface roughness and input process parameters, and then optimizing the output parameters will result in reasonably accurate results in HSM. However, there are many problems associated with the optimization of large-scale problems, such as multimodality, dimensionality and differentiability. Conventional approaches, e.g. linear and dynamic programming, etc., are not efficient in solving non-linear objective functions. Therefore, to develop efficient and effective optimization techniques is a matter of great urgency. Research is being conducted in this field and nature-inspired meta-heuristics optimization approaches are considered better than the traditional techniques, and thus are prominently used. One of the most efficient and effective optimization techniques reported is the gravitational search algorithm (GSA), which was developed by Rashedi et al. [8] and is based upon the gravitational and motion laws of Newton, where by every particle is considered an agent in the universe. The particles attract each other with a force that is dependent on their masses and distances. An optimization method, such as teaching-learning-based optimization (TLBO), is also used for obtaining global solutions instead of local optimal solution for continuous non-linear functions. The TLBO method works on the philosophy of teaching and learning [9]. These methods may not offer the best solutions to all types of problems. Some of them yield their best results for specific cases and thus other methods result in optimum values for other types of problems. Therefore, it is important to consider a few soft computing techniques for applications in order to know which best suits the given problem. In this work, the structure of casting of aluminium silicon carbide is tested for uniformity of distribution, by using an electron microscope before machining. Experiments were conducted on this material using a CNC high-speed machine by varying four selected input parameters: cutting speed, feed, depth of cut and step over ratio. The structure of material is also tested, using an electron microscope after machining. The surface roughness is measured using Taylor Hobson surface roughness tester (Surtronic 3+) for each of the forty selected combinations of input parameters. The surface roughness tester has a traverse speed of 1 mm/s; the measurement unit is mm present by the dual in-line package (DIP) switch, deselected via the menu; its cut-off values are 0.25, 0.8, and 2.5 mm; its traverse length % Lambda +n Lambda (n = 1, 3, 5, 10 or 25.4+0.2 mm at 0.8 mm cut-off); its parameters are Ra Rq, Rz (DIN), Ry and Sm; its calculation time is less than reversible time or 2 s, which ever is longer. An empirical relationship between surface roughness and the input parameters is established using the MINITAB software package. The empirical equations thus obtained are optimized using GSM, TLBO, simulated annealing (SA) and genetic algorithm (GA) methods, and the results are compared. This paper is organized as follows: in addition to the introduction in this section, a brief review of literature is given in Section 1. Experimental set up and conduction of experiments is explained in Section 2, and the results obtained and analysis are explained in Section 3. The conclusions drawn are given in Section 4. 1 LITERATURE REVIEW The results of some of the researchers who worked on conventional machines and high-speed CNC machines by varying selected input parameters are presented in the following section. The literature includes only soft computing applications and meta-heuristic methods in the analysis of machining parameters for composite materials. 1.1 Composite Material Pathak et al. [4] presented the preparation of an aluminium silicon carbide composite. They have also given a detailed discussion about the microstructure and its different mechanical properties. Neelima Devi et al. [10] studied the characterization of an aluminium silicon carbide composite. In their paper, tensile strength experiments were conducted by varying mass fraction of SiC (5, 10, 15, and 20 %) with aluminium. Palanikumar and Karthikeyan [3] investigated the factors influencing the surface roughness on the machining of Al/SiC particulate composites, using tungsten carbide tool inserts (K10). Dabade et al. [11] studied the surface integrity as a function of process parameters and tool geometry by analysing cutting forces, surface finish, and microstructures of the machined surfaces on Al/SiC/10p and Al/SiC/30p composites using cubic boron nitride (CBN) inserts. Basheer et al. [12] presented a model to predict the surface roughness in precise machining of metal matrix composites, using PCD tools with respect to the size and volume of reinforcement, tool nose radius, feed rate, and the depth of cut. Very few of them worked with composites in HSM conditions and checked the stability of composite by testing the microstructure after machining. 1.2 Conventional Machining Some researchers used statistical methods for analysing the relationships between surface roughness and selected input parameters. Aluminium metal matrix composite was used by Seeman et al. [13] whereas LM25 Al/SiCp composite was used by Arokiadass et al. [2]. Both used surface roughness and tool wear as output variables in their studies. They used RSM for establishing the relationship between input and output variables and conducted statistical analysis. Seeman et al. used machining time (t), estimated flank wear (VBmax) and surface roughness (Ra) in addition to the three parameters used by most of the researchers. Other researchers used soft computing techniques for analysis. The application of the GA method is the most common method found in the literature. Bhushan et al. [14] studied the 7075 Al alloy SiC composites and applied the GA technique to optimize the machining parameter in turning operations. Rai et al. [15] studied the multi-tool milling and estimated the machining time by applying GA techniques, taking axial depth of cut, radial immersion, feed rate and spindle speed as input parameters. Del Prete et al. [16] studied the flat end-milling process with feed, depth of cut, radial engage and speed as input parameters. Xu et al. [17] estimated various output parameters, such as cutting force, tool life and machined surface roughness by taking feed rate, depth of cut and cutting width as input parameters. Saffar and Razfar [18] estimated cutting force in the end-milling process by taking speed, feed rate and radial rake angle as input parameters. Alam et al. [19] predicted the surface roughness by considering speed, feed rate, and depth of cut as input parameters. They applied both the GA and quadratic prediction models to optimize the machining process parameters for the minimum surface roughness. Rao et al. [9] proposed TLBO for the optimization of mechanical design problems. In their paper, he compared this technique with other techniques and proved TLBO to be best. Rao et al. [5] also used TLBO techniques for non-linear large scale problems and compared the results with those obtained by other techniques such as GA, particle swarm optimization (PSO) and artificial bee colony (ABC) techniques, and obtained the best results. There is no specific heuristic algorithm suitable for all types of problems. The methods used may result in better solutions under specific conditions. Li et al. [20] developed models to estimate the cutting force, tool life and surface roughness. They applied PSO to improve the result. Pare et al. [21] used speed, feed, the depth of cut and the step-over ratio as input variables and optimized surface roughness using PSO. Unlike PSO, in which the two best positions decide the direction of search, GSA uses the weight of all agents for finding new search directions. Zubaidi et al. [22] applied the GSA for the optimization of cutting conditions for end-milling Ti6A14V alloy and obtained results that were proved to be better in comparison to GA and PSO techniques. In addition to the above application of heuristic search-based algorithms, some researchers combined selected algorithms (hybrid) and made attempts to apply them in different contexts. Farahnakian et al. [23] studied end-milling with PSO combined with a neural network (NN) algorithm to predict surface roughness and cutting forces. Cus and Zuperl [1] compared the PSO, GA and SA techniques to estimate the cutting force, the best results were obtained with PSO. Huang et al. [24] studied end-milling process and deployed PSO to minimize tool wear. They took speed, feed rate and width and observed that the memetic PSO (MPSO) algorithm has better performance than back-propagation NN, conventional wavelet neural networks (WNN) and GA-based WNN. Other works in this area include Zain et al. [25], who studied three parameters of end-milling for minimizing surface roughness. Based on the experimental data, it was concluded that process parameters should be set at the highest cutting speed and the rake angle with the lowest feed in order to obtain the minimum surface roughness. In the area of the face-milling operation, Back et al. [26] studied the effects of the insert run-out errors and the variation of the feed rate on the surface roughness. Thambu et al. [27] observed the machining of die cast Al alloy-SiC composites. Machining studies were conducted on the AlSiC composite work pieces with high-speed steel (HSS) end-mill tools in a milling machine at different speeds and feeds. Dweiri et al. [28] studied the down-milling machining process of Alumic-79 with an adaptive neuro fuzzy inference system to estimate the effect of machining variables, i.e. spindle speed, feed rate, depth of cut and number of flutes on the surface finish. Fuht and Wu [29] analysed the effect produced by the tool geometries and cutting conditions on the machined surface quality and developed a model predicting the surface quality for aluminium. 1.3 High Speed Machining Most of the researchers used statistical analysis methods in the area of high speed machining. They considered different sets of input parameters, output parameters and tried various analysis techniques based on statistics, including structured equation modelling (SEM) analysis, robust design, and Taguchi design approaches. Kauppinen [30] observed the results of several projects on high-speed milling for testing the suitability of high-speed milling for different kinds of materials. The materials machined were steels, aluminium alloys, graphites and polymer matrix composites. It is observed that most of the materials can be machined by high-speed milling. Pare et al. [31] conducted experiments varying the speed from 2500 to 7500 rpm on composites. They developed predictive models for surface roughness and used GSA for optimization. In order to determine whether there is uniform variation of surface roughness over speed variation, the range of 5000 to 7500 rpm was taken into consideration. TLBO [32] was used, and the results were studied. Koshy et al. [33] studied high-speed end-milling of hardened AISI D2 tool steel (58HRC) and estimated surface roughness and tool wear. Ozcelik and Bayramoglu [6] considered the step-over ratio in addition to the three cutting variables and developed a surface roughness prediction model for wet machining conditions. De Souza Jr. et al. [34] examined two face-milling cutter systems in the high-speed cutting of grey cast iron under a determined cutting condition. El-Wahab and Kishawy [35] deployed mathematical models to enhance the surface quality during CNC machining. Axinte and Dewes [36] observed the experimental results and related empirical models for work piece surface integrity (SI) during high-speed machining. Ghani et al. [37] implemented Taguchi optimization methodology to optimize cutting parameters, e.g. speed, feed and depth of cut, in end-milling during the machining of hardened steel with a TiN-coated carbide insert tool considering semi-finishing and finishing conditions of high-speed milling. The milling parameters used are cutting speed, feed rate and depth of cut. From the analysis of the result, the optimal combination of process parameters for the resultant low cutting force and good surface finish was determined. Arokiadass et al. [2] studied the machining properties of LM25 Al/SiCp composite in high-speed end-milling. They deployed the RSM and prepared a mathematical model to estimate the high-speed end-milling process parameter. Some researchers used either soft computing techniques or meta-heuristic techniques for the development of relationships among selected output and input variables. Yong et al. [38] studied highspeed milling processes with the axial depth of cut, radial depth of cut and helical angle as input parameters. They applied GA to estimate cutting force and MRR and achieved very good results. Öktem et al. [7] developed a methodology in order to determine the optimum cutting conditions leading to minimum surface roughness in the milling of mould surfaces, by coupling RSM with a developed GA. RSM was utilized to create an efficient analytical model for surface roughness in terms of cutting parameters: feed, cutting speed, axial depth of cut, radial depth of cut and machining tolerance. Therefore, there is a need to study in detail how surface roughness varies in wider speed ranges in the case of HSM. There is scope for computing the best machining conditions under high-speed machining for composite materials and applying new heuristic techniques, such as TLBO and GSM, in the area in order to obtain better results. The experimental procedure and results and analysis are explained in subsequent sections. 2 EXPERIMENTAL PROCEDURE The experimental procedure involves four stages: (i) the selection of proper combination of composite material, necessary cutting tools, and required equipment, (ii) data collection by conducting experiments as per the defined plan, (iii) the establishment of a surface roughness prediction model, and (iv) optimization of the predictive model for better results. Each of the stages is explained in the following sub-sections. 2.1 Selection of Proper Combination of Composite Material, Cutting Tools and Equipment The experimental study was carried out in dry cutting conditions on a high-speed CNC vertical milling machine (Model-Agni-BMV-45-T-20-Year 2008). Others features are a maximum spindle speed of 10,000 rpm, clamping area of 450x900 mm2, feed rate of 1 to 1000 ^m/rev, 15-kW drive motor with a table size of 800x500x550 mm3. The work piece material used was aluminium silicon carbide composite in the form of a 55x55x22 mm3 block. The gravity die-casting method was used for the preparation of Al-SiC material. The hardness of the composite material increased with the increase in the percentage of SiC in the composite. The percentage of constituents is selected based on experimentation with various combinations of Al and SiC and testing the mechanical properties of component. Trial machining is done and the microstructure is also tested before and after trail machining. The material properties of aluminium silicon carbide material are shown in Tables 1 and 2. A flat end mill (25 mm diameter, 45 1 helix angle, 4-flutes) produced by Addison was used for machining. Table 1. Chemical composition of (wt %) Al SiC Al SIC Fe Co 84.5 15.45 0.03 0.02 Table 2. composite Mechanical properties of aluminium silicon carbide Work material Tensile strength Elongation Density Hardness Al SiC composite 145.32 N/mm2 5.5 % 2779 kg/m3 63.8 BHN 2.1.1 Microstructure Analysis Testing of properties of Al-SiC before machining and micro-structure analysis is one of the methods to study the properties of composites material. A scanning electron microscope is used to observe the microstructure of composite material; after this analysis, various materials used in the composite can easily be observed. To perform the micro-structure analysis, a sample has been cut as per the requirement. Different degrees of fine papers were used to rub the sample starts with 600, 800, 1000, 1200 and 1500 degrees. Later, it was again rubbed, using a velvet cloth on polishing machine to finally check the microstructure. Etching solution (a mixture of 10 g NaOH, 5 g of K4 (Fe (CN)6) and 100 ml distilled water) was used to observe the microstructure of composite. Fig. 1. Micro structure of aluminium silicon carbide composite This microstructure showed mixing of aluminium and silicon carbide particles in uniform way as shown in Fig. 1. In the microstructure, if particles are segregated near grain boundaries it ideally means that the composite has high hardness; if it is away from grain boundaries, it means it has less hardness. In this case, many particles are going away from grain a boundary, which means that it is highly machinable and very useful for industrial purposes, especially in the aerospace and satellite industries. 2.1.2 Surface Roughness Measurement A portable Surtronic 3+ was used to measure the surface roughness (shown in Fig. 2). Four readings have been taken in the traverse direction. In case there is a large deviation, extreme values are discarded and additional readings have been taken so that the total number of readings is four. Fig. 2. Surface roughness measurement using Surtronic 3+ The average value of surface roughness is calculated and recorded. In this study, Ra values were obtained in the range of 1.25 to 3.50 mm. This variation is significant for aluminium composites in HSM for applications in the aerospace industry. The repeatability of the measurements was found to be in the range of 2 to 5 %, which was considered satisfactory for generating empirical models using the variables spindle speed, feed rate, depth of cut and step-over ratio. 2.2 Collection of Data by Conducting Experiments as per the Plan In this study, experimental work was conducted with aluminium silicon carbide composite material on a high-speed CNC milling machine. Five different values of cutting speeds (2000, 3000, 4000, 5000 and 6000 rpm) two values for the feed ranging from 100 and 200 дт/rev, two values of depth of cut (0.2 and 0.4 mm), and two values of the step-over ratio (50 and 60 %), have been taken to conduct the experiments. In this way (5x2x2x2), a total of 40 experiments were designed and conducted in this study and the data is tabulated in Table 5. Five different values are taken for cutting speed, as the range of speed in CNC HSM is very wide. The experiments conducted beyond the 6000 rpm speed range are not presented in this paper, as the variation in surface roughness obtained is marginal. 2.3 Establishment of Predicted Model for the Selected Output Measure A prediction model for surface roughness is established by taking surface roughness as a dependent variable and input variables (cutting speed, depth of cut, feed and step-over ratio) as independent variables. A regression relationship is established between them, and the statistical tests (F-test and /-tests) are carried out to test the significance levels. Both linear regression and non-linear regression are attempted. In linear regression analysis, the general form of equation is: Ra = a1x1 + a2x2 + a3x3 + a4x4 + K. (1) Similarly, general form of non-linear equation is: Ra = K x x^1 x x^2 X x3 X xa4, (2) By taking logarithms on both sides: logw(Ra) = ai xlogw(xi) + a2xlogw(x2) + + a3xlogw(x3) + a4xlogw(x4) + logw(K). (3) In both the above cases, a1, a2, a3, and a4 are parameters and x1, x2, x3, and x4 representthe selected input variables of cutting speed, feed, the depth of cut, and the step-over ratio, respectively. K denotes a constant. 2.4 Optimization of Predictive Model The ultimate objective of the optimization process for any problem is to assign values to a set of variables that will result in the best performance of a system. These setting variables can be calculated through deterministic or non-deterministic methods. In the deterministic method, a bottleneck is a local minima and in this regard artificial intelligence optimizing techniques are a better option. The predictive model obtained in the above subsection is optimized by using four meta-heuristic algorithms (GSA, TLBO, SA and GA) and is explained in the following sub-section as other algorithms are available in the literature. The program code for GSA is developed using MATLAB 2012. 2.4.1 Gravitational Search Algorithm The concept of algorithm is as follows: the performance of the agents that are considered as the objects is calculated by their masses. As in space, all these objects are attracted to each other because of gravitational force, and a shifting of smaller masses towards heavier masses takes place. The heavier masses, due to weight, change their positions more slowly in comparison to lighter ones. There can be four classifications for the masses: active, passive, inertial mass and position. A correlation can be established as the position of the masses, and the solution of the problem is similar. A fitness function is applied to determine its gravitational and inertial masses. The GSA is governed by two laws: (i) the law of gravity, and (ii) the law of motion. According to law of gravity, every point mass in the universe attracts every other point mass with a force that is directly proportional to the product of their masses and inversely proportional to the square of the distance between them [8]. The law of motion states that the current velocity of any mass will be equal to the sum of the fraction of its previous velocity and the variation within the velocity. Variation in the velocity of any mass is equal to the force acted on the system divided by mass of inertia. The different steps of the GSA are as follows: • The first population is initialized to start. • The fitness of each agent is evaluated and values are recorded. • G is updated and the best and worst fitness rates in the population for i = 1, 2,..., N are selected. • M (weighting in range [0, 1]) is updated. Then update Kbest and calculate F. Then update a. • The velocity and the position are updated: All positions should be in a feasible range. • Repeat steps 2 to 5 until the stop criteria. • End. The following parameters must be set in GSA: • Population size, N: the number of agents. • Dimension, D: The control variables number. • Maximum iteration, T: number of iterations to stop the optimization process if there is no convergence. • Initial gravity, G0: initial value of gravity, G. GSA is a memory-less algorithm and very effective in most types of optimization problems. 3 RESULTS AND DISCUSSIONS The microstructure of the composite material of the selected combination showed no significant difference before and after machining, thus indicating the material is stable over the selected range of speed variation. The microstructure after machining is shown in Fig. 3. In this paper, linear and non-linear relationships were established between input and output parameters to predict surface roughness. Four heuristic methods (GA, SA, TLBO and GSA) are used for optimization. Fig. 3 Micro-structure of aluminium silicon carbide composite All these four techniques are fundamentally different from each other: GA is based upon the survival of fittest, SA is based upon annealing, TLBO is inspired by the teaching-learning method and GSA is inspired by natural phenomenon of gravitational law. It is important to compare the result of these techniques so as to select the best technique for this case. Table 3. Selected input parameter Parameter Smallest value Largest value Speed (x1) [m/min] 2000 6000 Feed (x2) [urn/rev] 100 200 Depth of cut (x3) [mm] 0.2 0.4 Step-over ratio (x4) 0.5 0.6 MATLAB 2012 was used for preparation of codes. The ranges taken for conducting experiments with the selected input variables are given in Table 3. The smallest value is taken as 2000 rpm for the cutting speed, as the lower speed values are covered in the case of conventional machines by earlier researchers for this case. Based on the selected range, experiments were conducted on a high-speed CNC machine, and data was collected by varying the cutting speed, feed, and depth of cut and step-over ratio while machining the Al-SiC composite material. In this experiment, coolant was not used and other indirect factors (machine vibration, machine tolerance, and operator performance) are assumed to be insignificant. The combinations of input variables and the measured output variable are given in Table 4. Table 4. Experimental data S. no X1 x2 X3 X4 Ra S. no X1 X2 X3 X4 Ra 1 2000 100 0.2 0.5 2.30 21 4000 200 0.2 0.5 2.10 2 2000 100 0.2 0.6 3.03 22 4000 200 0.2 0.6 2.46 3 2000 100 0.4 0.5 2.62 23 4000 200 0.4 0.5 2.34 4 2000 100 0.4 0.6 3.13 24 4000 200 0.4 0.6 2.65 5 2000 200 0.2 0.5 2.65 25 5000 100 0.2 0.5 1.45 6 2000 200 0.2 0.6 3.18 26 5000 100 0.2 0.6 1.63 7 2000 200 0.4 0.5 2.95 27 5000 100 0.4 0.5 1.51 8 2000 200 0.4 0.6 3.25 28 5000 100 0.4 0.6 1.79 9 3000 100 0.2 0.5 2.15 29 5000 200 0.2 0.5 1.48 10 3000 100 0.2 0.6 2.51 30 5000 200 0.2 0.6 1.60 11 3000 100 0.4 0.5 2.45 31 5000 200 0.4 0.5 1.88 12 3000 100 0.4 0.6 2.90 32 5000 200 0.4 0.6 1.99 13 3000 200 0.2 0.5 2.26 33 6000 100 0.2 0.5 1.62 14 3000 200 0.2 0.6 2.63 34 6000 100 0.2 0.6 1.69 15 3000 200 0.4 0.5 2.54 35 6000 100 0.4 0.5 1.74 16 3000 200 0.4 0.6 3.02 36 6000 100 0.4 0.6 2.10 17 4000 100 0.2 0.5 1.98 37 6000 200 0.2 0.5 1.91 18 4000 100 0.2 0.6 2.24 38 6000 200 0.2 0.6 2.19 19 4000 100 0.4 0.5 2.20 39 6000 200 0.4 0.5 1.93 20 4000 100 0.4 0.6 2.54 40 6000 200 0.4 0.6 2.30 The regression equation, which relates the dependent variable, surface roughness and independent variables as given in Table 1, is given in Eq. (4). Ra = 0.893 - 0.00028 x1 + 0.00186 x2 + + 1.19 x3 + 3.39 x4 . (4) In this analysis, p values are close to zero and the F value is 43.77 which are considered a good fit between the dependent and independent variables. Similarly, with the help of MINITAB software based on the same 40 experimental data points, a non-linear regression equation was developed, given in Eq. (5). 3.1 Linear Regression Analysis For linear regression analysis and an associated statistical test to calculate a p value, for the coefficients associated with each independent variable, the MINITAB software package is used. logw(Ra) = 1.99 - 0.454 logw(x\) + 0.124 log^) + + 0.157 log^) + 0.794 log^). (5) Table 5a. Results of test of significance of independent variables - linear regression Predictor Coef SE Coef t P Constant 0.8925 0.4195 2.13 0.041 v -.0002799 0.0000244 -11.48 0.00 f .0018650 0.0006897 2.70 .011 d 1.1925 0.3449 3.46 .001 sr 3.3850 0.6897 4.91 0.00 S = 0.218111 R - - Sq = 83.3 % R - Sq(adj) = 81.4 % Table 5b. Results of analysis of variance Source DF SS MSE F value P value Regression 4 8.3289 2.0822 43.77 0.00 Residual error 35 1.6650 0.0476 Total 39 9.9939 In Table 5a, the associated variable is an effective and efficient predictor. The R2 value in statics also is key parameter because it indicates model performance. In this case, the R2 value is 81.4 % and this model explains 81.4 % of the variation in the dependent variable. In Table 5b, residual errors are the unexplained portion of the dependent variable; a large residual error means model is unfit but in this case residual error is not too much so this model is fit for prediction. The results suggest that all the four input parameters have significant effects on surface roughness. 3.2 Non-Linear Regression Analysis In this study, non-linear regression analysis is performed with experimental data. In Table 6, the test of significance and variance, p value is 0.01, so it is statistically significant at a 99 % confidence level, and the associated variable is an effective and efficient predictor. Standard error estimates are very low for all four parameters, which indicates that this model is fit for prediction of surface roughness. The coefficients of non-linear regression equation clearly indicate that the effect of all the process parameters is significant on the surface roughness. Table 6a. Results of test of significance -non-linear regression Predictor Coef SE Coef t P Constant 1.9944 0.1876 10.63 0.000 v -0.4542 0.04174 -10.88 0.000 f 0.1242 0.04673 2.66 0.012 d 0.1569 0.04673 3.36 0.002 sr 0.7940 0.1776 4.47 0.000 S = 0.0444809 R - Sq = 81.7 % R - Sq(adj) = 79.7 % Table 6b. Results of analysis of variance - linear regression Source DF SS MSE F value P value Regre-ssion 4 0.310046 0.0775 39.18 0.00 Residual error 35 0.069249 0.00198 Total 39 0.379295 3.3 Optimization Using Selected Heuristic Methods After developing the relationships between dependent and independent variables using linear and non-linear regression methods, four optimization techniques were used, as explained earlier, and the results are given in Tables 7 and 8 for linear and non-linear relations, respectively. The results obtained using the four selected techniques reveal that higher speed is desirable in order to obtain a minimum surface roughness value. In the case of feed rate, the values varied from 101.56 to 192.3 дт/rev; only one value was close to 200 and other three values were close to 100 дт/rev, indicating that the low value of feed is desirable for minimum surface roughness. In the case of depth of cut, three values are close to 0.2 mm, indicating that a low value of depth of cut is desirable for minimum surface roughness. The step-over ratio varies from 0.5 to 0.6, three values are close to 0.5, and one value is close to 0.6. As the number of iterations is greater in the case of SA and GSA, it may be concluded that a step-over ratio of 0.5 is desirable for further work. Table 7. Comparison of four selected optimization methods - linear regression (LR) Parameter GA SA TLBO GSA 5702.37 2238.07 5950.2 5739.88 X2 147.026 101.564 192.3 105.357 X3 0.24049 0.2 0.3 0.237 X4 0.5067 0.6 0.5 0.5224 Ra M 1.57 2.389 1.63 1.53 Iteration 51 6307 5 1000 Table 8. Comparison of four selected optimization methods - nonlinear regression (NLR) Parameter GA SA TLBO GSA x1 5127.23 2846.235 4677.3 5984.9 X2 102.467 102.434 161.2 124.71 X3 0.2 0.2 0.3 0.216 X4 0.519 .5 0.5 0.5074 Ra M 1.66 2.09 1.88 1.571 Iteration 59 5953 5 1000 From the analysis for non-linear relationship, it was observed from the Table 8 that cutting speed varies from 2846.24 to 5984.9 rpm, indicating that higher speed is desirable for minimum surface roughness. Similarly, a lower value of feed, a lower value of depth of cut and a 50 % step-over ratio is required for minimum surface roughness. From Tables 7 and 8, GSA gives the minimum value of surface roughness as compared to other selected techniques. Therefore, it may be concluded that GSA is very effective in comparison to other techniques. 3.4 Validation of Results For validation purposes, four separate experiments were conducted, and the data were recorded. The value of surface roughness is computed using Eqs. (4) and (5). The results are presented in Tables 9 and 10. It is observed that the maximum percentage of error was -12.0 % and thus the results are validated. From results, it is clear that the non-linear regression equation better describes the data, as the range of percentage error is 10.1 as compared to range of percentage error of 16.01 % in the case of linear regression relation. Therefore, Eq. (5) fits the data better as far as validation results and best fit tests are concerned. Table 9. Validation of results for LR Cd TD te Difference (Ra- - Raa ) 0 0 X1 X2 X3 X4 ! of ce 'S & E ce ce" X ^ 2 LLJ 3000 150 0.4 0.6 2.82 2.842 0.022 0.7 4000 150 0.4 0.5 2.14 2.223 0.0.083 3.73 6000 150 0.2 0.6 1.60 1.425 -0.175 -12.28 6000 150 0.4 0.5 1.75 1.663 -0.090155 -5.2 Table 10. Validation of results for NLR ed TD te Difference (Ra- - Raa ) 0 0 X1 X2 X3 X4 ce ce 'S & E Q^ ce" X ^ 0 ' ' LLJ 3000 150 0.4 0.6 2.82 2.77 -0.05 -1.8 4000 150 0.4 0.5 2.14 2.10 -0.04 -1.9 6000 150 0.2 0.6 1.60 1.814 +0.214 11.7 6000 150 0.4 0.5 1.75 1.75 0.00 0 4 CONCLUSIONS In this research, the microstructure of Al SiC composite is found to be stable after testing it before and after machining in HSM. The surface roughness models after optimization using four meta-heuristic techniques reveal that GSA proved to be the best among the four techniques in terms of optimum value of surface roughness and that TLBO gave better result in terms of number of iterations. The non-linear regression equation better explains the relationship between surface roughness and input parameters. This study can be a basis for future researchers as more input parameters need to be taken into consideration in order to predict the surface roughness at much higher speeds, i.e. of 6000 rpm onwards. The interaction effects, if any, can also be studied by future researchers. There is a scope to consider more such composite materials with various combinations of constituent materials under high-speed cutting conditions in order to compute the optimum values of input machining conditions for better design of manufacturing processes. 5 REFERENCES [1] Cus, F., Zuperl. (2007). Particle swarm intelligence based optimization of high speed end milling. Journal of Achievements in Materials and Manufacturing Engineering, vol. 22, no. 2, p. 75-78. [2] Arokiadass, R., Palaniradja, K., Alagumoorthi, N. (2012). Prediction and optimization of end milling process parameters of cast aluminium based MMC. Transactions of Nonferrous Metals Society of China, vol. 22, no. 7, p. 1568-1574, DOI:10.1016/S1003-6326(11)61357-5. [3] Palanikumar, K., Karthikeyan, R. (2007). Assessment of factors influencing surface roughness on the machining of Al/ SiC particulate composites. Materials and Design, vol. 28, no. 5, p. 1584-1591, D0I:10.1016/j.matdes.2006.02.010. [4] Pathak, J.P., Singh, J.K., Mohan, S. (2006). Sythensis and characterization of aluminium silicon carbide composite. Indian Journal of Engineering & Material Science, vol. 13, no. 3, p. 238-246. [5] Rao, R.V., Savsani, V.J., Vakharia, D.P. (2012). Teaching-learning based optimization: an optimization method for continuous non-linear large scale problems. Information Science, vol. 183, no. 1, p. 1-15, DOI:10.1016/j. ins.2011.08.006. [6] Ozcelik, B., Bayramoglu, M. (2006). The statistical modeling of surface roughness in high-speed flat end milling. International Journal of Machine Tools & Manufacturing, vol. 46, no. 12-13, p. 1395-1402, DOI:10.1016/j.ijmachtools.2005.10.005. [7] Öktem, H., Erzurumlu, T., Kurtaran, H. (2005). Application of response surface methodology in the optimization of cutting conditions for surface roughness. Journal of Material Processing Technology, vol. 170, no. 1-2, p. 11-16, DOI:10.1016/j.jmatprotec.2005.04.096. [8] Rashedi, E., Nezamabadi-pour, H., Saryazdi, S. (2009). GSA: A gravitational search algorithm. Information Sciences, vol. 179, no.13, p. 2232-2248, DOI:10.1016/j.ins.2009.03.004. [9] Rao, R.V., Savsani V.J., Vakharia, D.P. (2011). Teaching-learning-based optimization: A novel method for constrained mechanical design optimization problems. Computer Aided Design, vol. 43, no. 3, p. 303-315, DOI:10.1016/j. cad.2010.12.015. [10] Neelima Devi, C., Mahesh, V., Selvaraj, N. (2011). Mechanical characterization of aluminium silicon carbide composite. International Journal of Applied Engineering Research, vol. 1, no. 4, p. 793-799. [11] Dabade, U.V., Joshi, S.S., Balasubramaniam, R., Bhanuprasad, V.V. (2007). Surface finish and integrity of machined surfaces on Al/SiCp composites. Journal of Materials Processing Technology, vol. 192-193, p. 166-174, DOI:10.1016/j. jmatprotec.2007.04.044. [12] Basheer, A.C., Dabade, U.V., Joshi, S.S. (2008). Modeling of surface roughness in precision machining of metal matrix composites using ANN. Journal of Materials Processing Technology, vol. 197, no. 1-3, p. 439-444, DOI:10.1016/j. jmatprotec.2007.04.121. [13] Seeman, M., Ganesan, G., Karthikeyan, R., Velayudham, A. (2010). Study on tool wear and surface roughness in machining of particulate aluminium metal matrix composite-response surface methodology approach. The International Journal of Advanced Manufacturing Technology, vol. 48, no. 5-8, p. 613-624, DOI:10.1007/s00170-009-2297-z. [14] Bhushan, R.K., Kumar, S., Das, S. (2012). GA approach for optimization of surface roughness parameters in machining of Al alloy SiC particle composite. Journal of Materials Engineering and Performance, vol. 21, no. 8, p. 1676-1686, DOI:10.1007/s1665-011-0066-2. [15] Rai, J.K., Brand, D., Slama, M., Xirouchakis, P. (2011). Optimal selection of cutting parameters in multi-tool milling operations using a genetic algorithm. International Journal of Production Research, vol. 49, no. 10, p. 3045-3068, DOI:10.1080/00207540903382873. [16] Del Prete, A., De Vitis, A.A., Anglani, A. (2010). Roughness improvement in machining operations through coupled meta model and genetic algorithms technique. International Journal of Material Forming, vol. 3, no. 1, p. 467-470, DOI:10.1007/ s12289-010-0808-y. [17] Xu, F., Zhu, J.J., Wu, X., Zang, X.J., Zuo, D. W. (2010). Parameter optimization of milling Ti6Al4V using GA approach. Key Engineering Materials, vol. 426-427, p. 1-4, DOI:10.4028/ www.scientific.net/KEM.426-427.1. [18] Saffar, R.J., Razfar, M.R. (2010). Simulation of end milling operation for predicting cutting forces to minimize tool deflection by genetic algorithm. Machining Science and Technology, vol. 14, no. 1, p. 81-101, DOI:10.1080/10910340903586483. [19] Alam, S., Nurul Amin, A.K.M., Patwari, A. U., Konneh, M. (2010). Prediction and investigation of surface response in high speed end milling of Ti-6Al-4V and optimization by genetic algorithm. Advanced Materials Research, vol. 83-86, p. 10091015, DOI:10.4028/www.scientific.net/AMR.83-86.1009. [20] Li, J.G., Yao, Y.X., Gao, D., Liu, C.Q., Yuan, Z.J. (2008). Cutting parameters optimization by using particle swarm optimization (PSO). Applied Mechanics and Materials, vol. 10-12, p. 879883. [21] Pare, V., Agnihotri, G., Krishna, C.M. (2011). Optimization of cutting condition in end milling process with the approach of particle swarm optimization. International Journal of Mechanical and Industrial Engineering, vol. 1, no. 2, p. 22316477. [22] Zubaidi, S.A., Ghani, J.A., Haron, C.H.C. (2013). Optimization of cutting conditions for end milling of Ti6Al4V alloy by using a gravitational search algorithm (GSA). Meccanica, vol. 48, no. 7, p. 1701-1715, DOI:10.1007/s11012-013-9702-2. [23] Farahnakian, M., Razfar, M.R., Moghri, M., Asadnia, M. (2011). The selection of milling parameters by the PSO-based neural network modeling method. The International Journal of Advanced Manufacturing Technology, vol. 57, no. 1-4, p. 4960, DOI:10.1007/s00170-011-3262-1. [24] Huang, H., Li, A., Lin, X. (2007).Application of PSO-based wavelet neural network in tool wear monitoring. Proceedings of the IEEE International Conference on Automation and Logistics, p. 2813-2817, DOI:10.1109/ICAL.2007.4339060. [25] Zain, A.M., Haron, H., Sharif, S. (2011). Integration of simulated annealing and genetic algorithm to estimate optimal solutions for minimizing surface roughness in end milling Ti-6AL-4V. International Journal of Computer Integrated Manufacturing, vol. 24, no. 6, p. 574-592, DOI:10.1080/095 1192X.2011.566629. [26] Baek, D.K., Ko, T.J., Kim, H.S. (2001). Optimization of feed rate in a face milling operation using a surface roughness model. International Journal of Machine Tools & Manufacturing, vol. 41, no. 3, p. 451-462, DOI:10.1016/S0890-6955(00)00039-0. [27] Thambu, S., Marimuthu, K. (2010). Machining studies of die cast aluminium alloy-silicon carbide composites. International Journal of Minerals, Metallurgy and Materials, vol. 17, no. 5, p. 648-653, DOI:10.1007/s12613-010-0369-6. [28] Dweiri, F., Al-Jarrah, M., Al Wedyan, H. (2003). Fuzzy surface roughness modeling of CNC down milling of Alumin-79. Journal of Materials Processing Technology, vol. 133, no. 3, p. 266-275, D0I:10.1016/S0924-0136(02)00847-6. [29] Fuht, K.H., Wu, C.F. (1995). A proposed statistical model for surface quality prediction in end-milling of Al alloy. International Journal of Machine Tools and Manufacturing, vol. 35, no. 8, p. 1187-1200, D0I:10.1016/0890-6955(95)90408-E. [30] Kauppinen, V. (2004). High speed milling - a new manufacturing technology. 4th International DAAAM Conference on Industrial Engineering Innovation as Competitive Edge for SME, p. 131-134. [31] Pare, V., Agnihotri, G., Krishna, C.M. (2013). Optimization of machining parameters in high speed end milling of Al-SiC using gravitational search algorithm. International Journal of Recent Advances in Mechanical Engineering, vol. 2, no. 4, p. 45-51. [32] Pare, V., Agnihotri, G., Krishna, C.M. (2013). Optimization of machining parameters in high speed end milling of Al-SiC composites using teaching learning based optimization. Proceedings of the 2nd International Conference on Intelligent Robotics, Automation and Manufacturing, p. 522-528. [33] Koshy, P., Dewes, R.C., Aspinwall, D.K. (2002). High speed end milling of hardened AISI D2 tool steel (58 HRC). Journal of Materials Processing Technology, vol. 127, no. 2, p. 266-273, DOI:10.1016/S0924-0136(02)00155-3. [34] De Souza Jr., A.M., Sales, W.F., Santos, S.C., Machado, A.R. (2005). Performance of single Si3N4 and mixed Si3N4CPCBN wiper cutting tools applied to high speed face milling of cast iron. International Journal of Machine Tools & Manufacture, vol. 45, no. 3, p. 335-344, DOI:10.1016/j. ijmachtools.2004.08.006. [35] El-Wahab, A.I., Kishawy, H.A. (2000). A new method to improve the surface quality during CNC machining. International Journal of Production Research, vol. 38, no. 16, p. 3711-3723, DOI:10.1080/00207540050175969. [36] Axinte, D.A., Dewes, R.C. (2002). Surface integrity of hot work tool steel after high speed milling-experimental data and empirical models. Journal of Materials Processing Technology, vol. 127, no. 3, p. 325-335, DOI:10.1016/S0924-0136(02)00282-0. [37] Ghani, J.A., Choudhury, I.A. Hassan, H.H., (2004). Application of Taguchi method in the optimization of milling of Al alloy. Journal of Materials Processing Technology, vol. 145, no. 1, p. 84-92, DOI:10.1016/S0924-0136(03)00865-3. [38] Yong, Z.P., Jun, Z., Xiuli, F., Xing, A. (2010). Optimization of surface roughness based on multi-linear regression model and genetic algorithm. Advanced Materials Research, vol. 97101, p. 3050-3054, DOI:10.4028/www.scientific.net/AMR.97-101.3050. Strojniški vestnik - Journal of Mechanical Engineering 61(2015)3, 187-195 © 2015 Journal of Mechanical Engineering. All rights reserved. D0l:10.5545/sv-jme.2014.2210 Original Scientific Paper Adaptive Fuzzy-PI Control for Active Front Steering System of Armoured Vehicles: Outer Loop Control Design for Firing On The Move System Zulkiffli Abd Kadir*1,2 - Saiful Amri Mazlan2 - Hairi Zamzuri2 - Khisbullah Hudha1 - Noor Hafizah Amer1 JNational Defense University of Malaysia, Faculty of Engineering, Malaysia 2Technology University of Malaysia, Malaysia-Japan International Institute of Technology, Malaysia An armoured vehicle tends to lose its dynamic mobility when firing on the move. This is due to the effect of the firing force that reacts at the centre of the weapon platform, which creates an unwanted yaw moment at the vehicle's centre of gravity. In order to enhance the mobility performance of the armoured vehicle, a control strategy, i.e. yaw rejection control, is designed and test on an armoured vehicle model. The purpose of the control strategy is to maintain the directional mobility of the armoured vehicle by providing a steering correction angle to the pitman arm steering system. The control strategy proposed in this study consists of two main structures: yaw rate feedback control using a Proportional-Integral-Derivative (PID) controller and Lateral Force Rejection Control (LFRC) using an adaptive Fuzzy-Proportional-Integral (adaptive Fuzzy-PI) controller. The simulation results in terms of yaw and lateral motions were observed, and the proposed control strategy was shown to successfully improve the directional mobility of the armoured vehicle after firing. The benefit of the proposed control strategy with adaptive fuzzy-PI control is evaluated by comparing its performance to fuzzy-PI and proportional-integral (PI) control strategies. Keywords: armoured vehicle model, validation of armoured vehicle model, active front steering, firing on the move, adaptive fuzzy control, lateral force control Highlights • The armoured vehicle model is developed. • The armoured vehicle model was validated with a real armoured vehicle. • Firing on the move control strategy is proposed. • The control strategy consists of adaptive fuzzy control with lateral force control. 0 INTRODUCTION An online controller is used to provide a stable condition for engineering systems due to the external disturbances and uncertain inputs by automatically changing the controller parameters to satisfy the necessity of system performance. Previous studies confirmed the advantages of fuzzy controller, which has intelligent functions that can be used to predict responses through a learning process, perform optimization work and is capable of maintaining the system performance due to changing condition [1]. However, there are several problems commonly associated with fuzzy controllers. First, it is always designed without obtaining the optimum range of the membership function inputs, value of outputs and controller rules. Without an optimum set of parameters, the performance of the controller and its ability to achieve a specific desired output will be compromised. The second common problem is developing the fuzzy controller on a nonlinear model without identifying the actual system parameters. In this case, a fuzzy controller will produce insignificant output performances. Hence, the control strategy using the existing fuzzy membership function will not be able to produce the optimum performance during the initial analysis work in order to achieve the required performance levels. Based on the abovementioned problems, a method to automatically adjust the fuzzy controller parameters is required to overcome the effect of external disturbances while dealing with parameter optimizations and nonlinear system models. The parameters known as the centre and spread of the membership function for both inputs and output values were adjusted and optimized automatically to improve the controller performance. Fuzzy controllers with optimization capabilities are often called as "adaptive fuzzy". Currently, there are some research works that have proposed adaptive fuzzy controllers for nonlinear vehicle systems. Most of the adaptive fuzzy controllers were developed to meet the desired system performance, such as improving vehicle handling and ride stabilities performance using active suspension [2] and [3], controlling vehicle slip using antilock braking system [4], monitoring the vehicle gap keeping [5], developing electrical system for electric vehicle [6] and [7] and yaw moment control for independent rear wheel drives [8]. The primary intention of the proposed adaptive fuzzy controllers was to solve the problem of selecting the range of controller parameters that can deliver the desired system performance. In this study, a control strategy of yaw rejection system for armoured vehicle using active front steering (AFS) system is proposed. The proposed control strategy for yaw rejection is developed on a validated four-wheeled armoured vehicle model. It consists of two outer loop controllers for yaw rate feedback and lateral force feedback. The aim of designing the outer loop controllers is to reduce the magnitude of unwanted motion in the lateral direction by minimizing yaw and lateral motions due to firing force. The outer loop controllers for yaw rate feedback and lateral force feedback proposed in this study have been developed using proportionalintegral-derivative (PID) control and fuzzy-PI control strategy, respectively. The proposed control strategy with adaptive fuzzy-proportional-integral (fuzzy-PI) control is then compared against the general fuzzy-PI control and the conventional proportional-integral (PI) control to evaluate the effectiveness of adaptive fuzzy-PI control. This paper is structured as follows: the introduction and previous works on adaptive fuzzy control are explained in the first section. The second section focuses on the mathematical equations of the armoured vehicle model and is followed by validation model in the third section. The next section shows the development of the proposed control strategy with adaptive Fuzzy-PI control. The sixth section discusses the simulation results of the adaptive Fuzzy-PI control and is followed by the overall conclusion in the last section. 1 ARMOURED VEHICLE MODEL The armoured vehicle considered in this study, as illustrated in Fig. 1, is based on a category of a movable weapon system with four wheels and can fire a projectile from the gun turret. heave as well as allowing movements in longitudinal, lateral and yaw directions at its centre of gravity (CG). Fig. 1. A 2D view of four wheel armoured vehicle The armoured vehicle covers the ride model and the handling model, which is free to pitch, roll and Fig. 2. Armoured vehicle ride model Based on the armoured vehicle ride model shown in Fig. 2, the dynamics behaviour of the armoured vehicle in terms of pitch, roll, and heave motions are derived mathematically using equations of motions from Newton's second law as follows: Ipe = ( Fsri + Fdrl + Fsrr + Fdrr )b - - (Fsfl + Fdfl + Fsfr + Fdfr )a + msaÄ (1) = (Fsfr + Fsrr + Fdfr + Fdrr ) ) - -(Fsfl + Fsrl + Fdf + Fdrl ) ) + mayhCG , (2) mZs = Fsfi + Fdfl + Fsfr + FdJr + FsrI + FdrI + Fsrr + Fdrr ,(3) where p is the angular acceleration of roll at body CG [rad/s2], Ir is the moment inertia of roll axis [kg/m2], hCG is the distance between roll axis and CG [m], w is the vehicle track width [m], ay is lateral acceleration [m/s2], Q is the angular acceleration of pitch at body CG [rad/s2], Ip is the moment inertia of pitch axis [kg/m2], a is distance from the front tyre to the body CG [m], b is distance from the rear tyre to the body CG [m], ax is longitudinal acceleration [m/s2], ms is the armoured vehicle's mass [kg], Zs is the vertical acceleration of sprung mass at CG [m/s2], Fsij is suspension force [N] and Fdj is damper force [N]. Fig. 3 illustrates the armoured vehicle handling model. The longitudinal and lateral accelerations can be obtained by considering the forces acting on each tyre in the longitudinal direction, x-axis and longitudinal direction, y-axis as follows: Fxrr + Fxrl - Fyfr sin 5 + Fxfr c0s5- -Fyfisin s + Ff cos ö - Fecos ß = msax, (4) yrl -Ff sm5 -Fyf, cosS- +Fxfl sin 5 + Fyfl COsS + Fe sin ß= msay> (5) Fig. 3. Armoured vehicle handling model with firing disturbance The yaw angular acceleration acting on the armoured vehicle handling model is also based on the effects of tyre forces, i.e. lateral force and longitudinal force, and it can be derived as: ( W 17 - J Fxfr cos w w -—Frfl cos S--Frr 2 xß 2 w w w +—F xri + у Fyß sin - у Fyfr sm - -bFyrl - bFyrr + aFyfl C0s ' + aFßr C0s ' -aFß sln ' - aFßr sln ' - cFe sln ß = Iz Г, (6) where Iz is the moment of inertia around z-axis [kg/m2], 3 is the tyre steer angle [rad], r is yaw angular acceleration [rad/s2] and ms is the armoured vehicle's mass [kg]. 2 VALIDATION OF ARMOURED VEHICLE MODEL In order to validate the armoured vehicle model, a handling test was conducted using a Ferret Scout Car armoured vehicle. Several transducers were installed in the armoured vehicle to observe the vehicle's behaviour such as lateral acceleration, roll angle, yaw rate and steering input from the driver, as shown in Fig. 4. Validation of the developed armoured vehicle model can be defined as an assessment and comparison of the developed model's behaviours with the actual armoured vehicle's behaviours [9]. The validation of the armoured vehicle model is done by comparing the behaviour obtained from the simulation model with the actual responses obtained from the experimental armoured vehicle using similar handling test. The experimental results were also used to optimize the parameters of the developed armoured vehicle model. Fig. 4. Instrumented experimental armored vehicle The model validation is performed for a double lane change test. However, due to several constraints, such as limitation of the experimental area and the speed of the Ferret Scout Car, the experimental work is performed with speed of 40 km/h. Fig. 5 shows the steering input delivered by the driver in double lane change test. Fig. 5. Steering wheel angle Fig. 6. Lateral acceleration response Fig. 7. Yaw rate response The validation results for the double lane change test are analysed in terms of lateral acceleration, yaw rate and roll angle of the armoured vehicle. The results are shown in Fig. 6 to 8, respectively. It can be seen that simulation model has similar trends and magnitudes with the actual manoeuvring of the armoured vehicle. The small deviations in the validation results were caused by the error from the driver in maintaining a constant speed for the actual handling test. For the overall validation results, it is clear that the behaviours obtained from the simulation work and the experimental work for the double lane change test have similar responses with acceptable rates of error. The validated armoured vehicle model is then used to develop a yaw cancellation system using active front steering in the next section. Fig. 8. Roll angle response 3 THE PROPOSED CONTROL STRATEGY OF AFS FOR ARMOURED VEHICLEs The AFS for four-wheeled armoured vehicles is developed by installing an additional mechanism consisting of a direct current (DC) motor, known as an AFS actuator as well as AFS gearbox, as illustrated in Fig. 9. SM* PITMAN ARM m Fig. 9. Configuration of active front steering using pitman steering This mechanism is installed in the conventional Pitman arm steering system by maintaining the existing parts, including the steering wheel, steering column, recirculating ball gearbox, pitman arm and steering linkage. The AFS gearbox is used to superimpose the steering wheel angle obtained from the driver and angle of steering correction from the AFS actuator. In AFS, there are two angles need to be evaluated: the steering wheel input given by the driver and the steering correction angle generated by the AFS actuator. The combination of both angles provides a correction angle to the front wheels through the Pitman arm joint. In this study, an additional controller loop is introduced as lateral force rejection control (LFRC). The purposed of LFRC loop is used to measure the estimated lateral force (Fye) in order to eliminate the unwanted lateral motion due to the firing impact. The LFRC is adopted from the principle of skyhook control usually used in suspension control, the main purpose of which is to improve the vertical behaviour of the vehicle subjected to road disturbances [10] and [11]. Using the same concept, skyhook control principle was applied in LFRC to reduce unwanted disturbances in lateral direction. The configuration of the estimated lateral force is shown in Fig. 10, which consists of an imaginary lateral damper (СШега[) to provide virtual damping for the system and an imaginary lateral wall as the reference point. Vehicle CG Firing force Fig. 10. Armored vehicle with imaginary lateral wall and imaginary lateral damper The estimated lateral force (Fye) is defined as: Fye = Claterali GC . (7) Here, yGC is known as the lateral velocity of the armoured vehicle. This can be obtained by differentiating the global coordinate of lateral displacement (yGC) of the armoured vehicle. The lateral displacement of yGC is defined as: yGC = Уо +J Vy cos r + vx sin r, (8) where yo is the local lateral displacement [m], r is the armoured vehicle's yaw angle [rad], vx the armoured vehicle's longitudinal velocity [m/s], and vy the armoured vehicle's lateral velocity [m/s]. The proposed AFS control strategy is designed using a four-wheel armoured vehicle model as shown in Fig. 11. The system is developed by assuming that the armoured vehicle travels in the longitudinal direction without any steering input from the driver. Fig. 11. Control strategy of AFS for armored vehicle This control strategy consists of two feedback loop controllers, which are inner and outer loop controls. The inner loop controller is developed to control the AFS actuator while the outer loop controller is to control the overall dynamic performance of the armoured vehicle during firing. In the outer loop controller, both the yaw rate and lateral force controller feedback are used to reject the unwanted yaw motion while simultaneously minimizing the lateral movement of the armoured vehicle subjected to firing impact. A conventional PID controller is used as the yaw rate feedback controller in order to estimate the yaw rate error of the armoured vehicle. This error is used as the reference point for the following controller loop, known as the lateral force control. This lateral force control is monitored by a self-tuning PI control adapted by a nonlinear fuzzy membership function. Henceforth, the desired lateral force (Fyd) is defined as: Fyd = (Kpy + K + Kdys)(yd - ya ), (9) where ya is actual yaw rate, yd is desired yaw rate, while Kpy, Kiy/S and Kdy s are vehicle controllers for yaw rate response. Fyd is compared with the estimated lateral force (Fye) in order to measure the desired angle of steering correction (Sd). The steering correction is used to counterback the steering wheel angle due to the firing disturbance acting at the centre of weapon platform. Therefore, the Sd is measured as below: = (Kpl + Kl)(Fyd - Fye ), (10) where Kpl and Kn/ S are proportional and integral gains for lateral force control, respectively. Initially, the gain values of proportional and integral are set to be a constant values throughout the simulation procedure. However, these constant values are not compatible once the conditions or disturbances vary with time. Hence, a self-tuning method is required in order to tune the constant values according to the disturbance or condition. The self-tuning method can be implemented for both proportional and integral gains by identifying the maximum and minimum gain values through the simulation procedure. The range of the proportional and integral gains are Kpl e [1.5, 0.1] and Ku e [0.08, 0.01]. The range of each gains can be used to calibrate K'pl and Kil as follows: -pi K — K K' _ pl pl min K — K pl max pl min Kp, — 0.1 1.5 — 0.1 ' K — K K' _ il il min K — K il max il min Ki, — 0.01 0.08 — 0.01' (11) (12) where K'pl and K'il are the update values for Kpl and Kil . Both equations can be rearranged as Kpl = 1.4 K'pl+0.1 and Kit=0.07 K'n+0.01. Although the self-tuning PI control is used in this control strategy, the performance of the armoured vehicle due to firing needs more improvement in term of lateral displacement and the stability of the armoured vehicle. Hence, a nonlinear fuzzy control is merged with a self-tuning PI controller. The combination of a self-tuning Fuzzy-PI controller is used to increase the dynamic performance of the armoured vehicle during firing. The equations of the nonlinear fuzzy control are described as follows: ^LA-П exp J, = f ( *|0) = - ( i \ xj - cj v a'j J КД -exp 2 v, v J J (13) where, x}- is input to the fuzzy set, j = 1, ..., n (n is number of inputs) and i = 1, ..., r (r is the number of rules). The parameter bi is the output values of the fuzzy set, while cj and oj are centre and spread of Gaussian membership function inputs, respectively. However, the parameters of the fuzzy control itself need to be configured varying with time since there are several constant parameters which are b, c and a. Therefore, an adaption law is absorbed in the self-tuning Fuzzy-PI controller in order to enhance the performance of the AFS control strategy. The adaptation law is developed based on an adjustment mechanism to update the constant parameters of the fuzzy controller. The adjustment mechanism for the fuzzy controller is developed using first order reference model as shown in Fig. 11. The intention of the adjustment mechanism is to vary the fuzzy controller parameters, i.e. b, c and a. The yaw rate (ya) response is considered in designing the adjustment mechanism of the fuzzy control. It is assumed that output from fuzzy control f(x\0) is similar with the yaw rate response, and error (em) for the adjustment mechanism can be written as: = Л - Ут )2 = f ( х\в) - ym ) (14) em is designed to minimize error between the reference model (ym) and yaw rate response (ya). By considering the number of rules (r), the parameters of input c, a and the output parameter b can be adjusted by updating the parameter values using gradient descent method as follows: b (k+1) = b (k ), dbt cip c (k +1) = c (k ), dc. (15) (16) de a i (k +1) = ai (k ) - A3 ——, da, (17) where, update parameter values A1, X2 and X3 need to be fine tuned. The first derivative of adjustment error (em) with respect to fuzzy parameters b, c and a are written as follows: b (k +1) = b (k) -V f (x | в) - ym ) -, (18) w c(k +1) = c(k) - Я2[ f (х\в) - ym ][b(k) - X - c'j (k) - f ( xm\e(k ))] (a' (k ))2 -, (19) w a, (k +1) = a,, (k ) - Яз[ f ( х\в) - ][bi (k ) - - f(xm \в(к))] (xX; - c) (k))2 (aj (k ))3 -, (20) w where, z' =П exp j=1 ( m Xj - cj У aJ J i\ and (21) (22) 4 PERFORMANCE ASSESSMENT OF THE ADAPTIVE FUZZY-PI CONTROL FOR ARMOURED VEHICLE USING ACTIVE FRONT STEERING The performance of proposed control strategy is analysed for a duration of 2.5 seconds. The control parameters of PI control, parameters of the armoured vehicle model, membership functions of Fuzzy-PI, input rules of Fuzzy-PI control output rules of fuzzy-PI control and control parameters of adaptive fuzzy-PI control are presented in Tables 1 and 2, Fig. 12 and Tables 3 to 5, respectively. Table 1. Controller parameters for control strategy with PI control Yaw Rate Controller K II 40 Kiy = 4 Kdy = 3 Lateral Force Controller Kpl = 1.5 Kil = 0.08 The cumulative Gaussian for error and error rate membership functions are set as negative big (NB), negative small (NS), zero (ZE), positive small (PS) and positive big (PB). Meanwhile, the output values =1 to update the PI control are set as big (B), medium big (MB), medium (M), medium small (MS) and small (S). Table 2. Armoured vehicle model parameters Armoured vehicle mass, ms [kg] Distance front tyre to the body CG, a [m] Distance rear tyre to the body CG, b [m] Distance from roll axis to CG, hCG [m] Moment of inertia around z-axis, Iz [kg/m2] Moment inertia of roll axis, Ir [kg/m2] Track width, w [m] a) Error membership function 4000 1.3 0.5 5000 500 Damping stiffness at front left, front right, rear left, rear right, Ksß, Кф, KsH, KSrr [N/m]_ 10000 Damping stiffness at front left, front right, rear left, rear right, Csß, Csfr, Csrh Csrr [N/ms1] 800 1.5 b) Error rate membership function Fig. 12. Membership functions of Fuzzy-PI control for error rate and error Table 3. Fuzzy rules for control strategy with Fuzzy-PI control Error Rate/Error NB NS ZE PS PB NB MS MS NS MS MS MB ZE MS MS MB MB PS MS MB MB PB MB MB Table 4. Fuzzy output values for control strategy with Fuzzy-PI control Update value for PI control S MS M MB B К' pi 0.1 0.75 1.5 2.25 3 К il 0.01 0.04 0.08 0.12 0.16 Table 5. Controller parameters for control strategy with adaptive Fuzzy-PI control Update parameter Value 0.05 ¥ ~ 0.04 cu I 0.03 » 0.02 ф 0.01 -30 4000 10000 ■■■PI control --fuzzy-PI control —adaptive fuzzy-PI control 0.5 1 1.5 Time [s] 2 2.5 a) Firing angle of 30 degrees b) Firing angle of 60 degrees Fig. 13 c) Firing angle of 90 degrees Lateral displacement responses for various firing angles Simulation results of armoured vehicle body displacement in lateral motion are shown in Fig. 13. The firing angles of the armoured vehicle are set at 30, 60 and 90 degrees with a constant cruising speed of 40 km/h. The effectiveness of the adaptive Fuzzy- ; S S и S M У У R У p R PI control strategy can be evaluated from the lateral displacement error of the armoured vehicle after the firing condition. The control strategy with adaptive Fuzzy-PI control managed to reduce the magnitude of lateral displacement error and maintain the direction of travel after firing, in comparison to its counterparts, i.e. Fuzzy-PI and PI controls. The controller's ability to improve the dynamics performance in lateral direction will ensure better mobility for the armoured vehicle when firing on the move. a) Firing angle of 30 degrees b) Firing angle of 60 degrees c) Firing angle of 90 degrees Fig. 14. Yaw rate responses for various firing angles Similar performance also can be observed in the yaw rate and yaw angle, as shown in Figs. 14 and 15, respectively. In both cases, the control strategy with the adaptive Fuzzy-PI control shows better performance in minimizing both the settling time and the response magnitude in comparison to the control strategy with Fuzzy-PI and PI controls. For the yaw rate response in Fig. 14, it can be observed that the control strategy with adaptive Fuzzy-PI control has been successful in minimizing the yaw rate magnitude as compared to its counterparts. The control strategy with the adaptive Fuzzy-PI control also managed to isolate armoured vehicle body from the unwanted yaw rate almost 20 % more than the Fuzzy-PI control for all firing angles. By isolating the unwanted yaw rate from the armoured vehicle body, the dynamics stability of the armoured vehicle during firing on the move can be improved. This will improve the vehicle's ability to move at the intended direction when firing. Furthermore, the probability of the armoured vehicle becoming unstable can be minimized. a) Firing angle of 30 degrees X 10 "3 « 2 PI control —fuzzy-PI control —adaptive fuzzy-PI control / / \ ; / л % //\\ 0.5 1 1.5 Time [s] b) Firing angle of 60 degrees 2.5 X 10 — ч -: О 1 ГО 52 ш >- 3I control —fuzzy-PI control / —adaptive fuzzy-PI control V ' 4 ■ / / \ V 0.5 1 1.5 Time [s] 2.5 c) Firing angle of 90 degrees Fig. 15. Yaw angle responses for various firing angles Similarly, the control strategy with the adaptive Fuzzy-PI control demonstrates a significant reduction on yaw angle response at body centre of gravity on firing test with angles of 30, 60 and 90 degrees compared to Fuzzy-PI and PI controls, as shown in Fig. 15. From the simulation results, it is evident that the adaptive Fuzzy-PI control is very effective in minimizing the yaw angle error. The yaw angle response for all firing tests is summarized in Table 6. By minimizing the yaw angle error, directional stability of armoured vehicle during firing while moving can be improved, which is the major aspect of armoured vehicle safety. Table 6. Yaw angle responses for firing angle of 30, 60 and 90 degrees Firing Yaw angle response angle [°] PI [rad] Fuzzy-PI [rad] Adaptive fuzzy-PI [rad] 30 2.5x10-4 1.5x10-4 1.1x10-4 60 4.6x10-4 2.9x10-4 2.2x10-4 90 5.4x10-4 3.6x10-4 2.9x10-4 5 CONCLUSION A control strategy for an armoured vehicle using AFS system has been developed on a validated model of a four-wheel armoured vehicle using MATLAB-SIMULINK. The main purpose of this proposed control strategy is to enhance the dynamic performance of the armoured vehicle, such as lateral displacement, yaw rate and yaw angle when firing on the move. There are two feedback controllers used in this study: yaw rate feedback using PID control and lateral force feedback using adaptive Fuzzy-PI control. The proposed AFS control strategy is then evaluated via simulation involving various firing angles of 30, 60 and 90 degrees. Based on the simulation results, it can be observed that the proposed AFS control strategy using the adaptive Fuzzy-PI controller performed significantly better in comparison to the Fuzzy-PI and conventional PI control strategy. This improvement of the armoured vehicle due to lateral and yaw performances can increase the stability of the armoured vehicle after the firing impact. Furthermore, the soldiers driving the armoured vehicle are able to control the vehicle even after the firing impact from the gun. 6 ACKNOWLEDGMENT The authors would like to thank the Malaysian Ministry of Education through RAGS (RAGS/2012/ UPNM/TK01/4), LRGS (LRGS/B-U/2013/UPNM/ DEFENCE & SECURITY-P1), Institut Kejuruteraan Tentera Darat (IJED) and the Malaysia-Japan International Institute of Technology (MJIIT) for their financial support and the use of their research facility for this research. 7 REFERENCES [1] Sa-ngawong, N., Ngamroo, I. (2015). Intelligent photovoltaic farms for robust frequency stabilization in multi-area interconnected power system based on PSO-based optimal Sugeno fuzzy logic control. Renewable Energy, vol. 74, p. 555567, D0l:10.1016/j.renene.2014.08.057. [2] Chiou J.S., Tsai, S.H., Liu, M.T. (2012). A PSO-based adaptive fuzzy PID-controllers. Simulation Modelling Practice and Theory, vol. 26, p. 49-59, D0I:10.1016/j.simpat. 2012.04.001. [3] Mailah, M., Priyandoko, G. (2007). Simulation of a suspension system with adaptive fuzzy active force control. International Journal of Simulation Modelling, vol. 6, no. 1, p. 25-36, DOI:10.2507/IJSIMM06(1)3.079. [4] Wang, W.Y., Li, I.H., Tsai, C.P., Su, S.F., Hsu, S.B. (2009). Dynamic Slip-Ratio Estimation and Control of Antilock Braking Systems Using an Observer-Based Direct Adaptive Fuzzy-Neural Controller. IEEE Transactions on Industrial Electronics, vol. 56, no. 5, p. 1746-1756, D0I:10.1109/TIE.2008.2009439. [5] Naranjo, J.E., Gonzalez, C., Reviejo, J., Garcia, R., de Pedro, T. (2003). Systems adaptive fuzzy control for inter-vehicle gap keeping. IEEE Transactions on Intelligent Transportation Systems, vol. 4, no. 3, p. 132-142, D0I:10.1109/ TITS.2003.821294. [6] Shen, W.X., Chan, C.C., Lo, E.W.C., Chau, K.T. (2002). Adaptive neuro-fuzzy modeling of battery residual capacity for electric vehicles. IEEE Transactions on Industrial Electronics, vol. 49, no. 3, p. 677-684, D0I:10.1109/TIE.2002.1005395. [7] Chau, K.T., Wu, K.C., Chan, C.C. (2004). A new battery capacity indicator for lithium-ion battery powered electric vehicles using adaptive neuro-fuzzy inference system. Energy Conversion and Management, vol. 45, no. 11-12, p. 1681-1692, D0I:10.1016/j. enconman.2003.09.031. [8] Niasar, A.H., Moghbeli, H., Kazemi, R. (2003). Yaw moment control via emotional Adaptive Neuro-Fuzzy Controller for independent rear wheel drives of an electric vehicle. Proceedings of2003 IEEE Conference on Control Applications, vol. 1, p. 380-385. [9] Kadir, Z.A., Hudha, K., Ahmad, F., Abdullah, M.F., Norwazan, A.R., Mohd Fazli, M.Y., Khalid, A.J., Gunasilan, M. (2012). Verification of 14DOF full vehicle model based on steering wheel input. Applied Mechanics and Materials, vol. 165, p. 109-113, D0I:10.4028/www.scientific.net/AMM. 165.109. [10] Harun, M.H., Abdullah,W., Jamaluddin, H., Rahman, R.A. Hudha, K. (2014). Hybrid skyhook-stability augmentation system for ride quality improvement of railway vehicle. Applied Mechanics and Materials, vol. 663, p. 141-145, D0I:10.4028/ www.scientific.net/ AMM.663.141. [11] Crews, J.H., Mattson, M.G., Buckner, G.D. (2011). Multi-objective control optimization for semi-active vehicle suspensions. Journal of Sound and Vibration, vol. 330, no. 23, p. 5502-5516, D0I:10.1016/j.jsv.2011.05.036. Strojniški vestnik - Journal of Mechanical Engineering 61(2015)3, 196-206 © 2015 Journal of Mechanical Engineering. All rights reserved. D0l:10.5545/sv-jme.2014.2138 Original Scientific Paper Study of an Energy Regeneration System with Accumulator for Hydraulic Impulse Testing Equipment Zaipeng Man* - Fan Ding - Chuan Ding - Shuo Liu Zhejiang University, The State Key Lab of Fluid Power Transmission and Control, China To guarantee the reliability and performance of hydraulic systems, it is of vital importance to do hydraulic impulse testing for hydraulic components. However, the testing time is usually so long that the energy consumption is very large. An energy regeneration system with an accumulator is proposed and used for hydraulic impulse testing equipment. Mathematical analysis and simulations show that the hydraulic impulse testing system with an accumulator can reduce energy consumption by 15% compared with that of the system without an accumulator in the cycle. Furthermore, the energy efficiency of the hydraulic impulse testing system is increased from 62.82 to 75.71% by the use of an accumulator. Both simulation results and experimental results show that the control performance of the hydraulic impulse testing equipment with an accumulator is better than that of the equipment without an accumulator, and has a faster response speed and a larger pressure rise rate. Keywords: hydraulic impulse testing, energy regeneration, hydraulic accumulator, simulation, experiment, energy storage Highlights • Proposed energy regeneration system with accumulator. • Energy regeneration system used for hydraulic impulse testing. • Mathematical and simulation models built for the system. • Effects on energy consumption and system efficiency studied. • Better control performance proved by simulation and experiment results. 0 INTRODUCTION Hydraulic transmission systems are widely used in construction and agricultural machinery owing to their high power density, high compactness, and flexible control [1]. However, with fuel prices increasing and an energy crisis looming, the large energy consumption of hydraulic systems is drawing more and more attention [2]. In order to improve the efficiency of hydraulic transmission systems and reduce energy consumption, many methods have been put forward. It is claimed that an energy-saving hydraulic system should be able to recover energy during deceleration or while a load is lowered [3]. Various energy-saving hydraulic systems have been studied, such as secondary control systems and constant pressure systems [4] to [6]. An energy regeneration system in hydraulic forklift trucks has been studied, concentrating on energy recovery in the main lift system with electric motor and batteries, and resulting in improved energy efficiency but shorter lifetime of components [7]. An energy recovery system with a hydraulic accumulator that could save and restore energy in a crane's hydraulic system has also been studied, and it was found that the potential energy of the crane and load can be saved in the form of hydraulic energy and reutilized [8]. A speed control system of a variable voltage variable frequency hydraulic elevator with a pressure accumulator was studied, and was shown to have higher efficiency compared with a hydraulic elevator without a pressure accumulator [9]. A new hydraulic closed-loop hydrostatic transmission energy-saving system has been proposed, and results indicate that the efficiency can be improved by using a hydraulic accumulator [3]. The energy regeneration systems for hybrid hydraulic excavators have also been studied, and it was found that an estimated 41% of the total potential energy could be regenerated at the lowering of the boom [10]. On the other hand, it is widely recognized that many hydraulic system accidents and component failures are caused by hydraulic impulse rather than overload [11]. So it is of vital importance to do hydraulic impulse testing for hydraulic components, including hydraulic hoses, fittings, assemblies, and so on. Although energy recovery applications are widely used in industry, few energy recovery systems have been developed for hydraulic impulse testing equipment. According to the relevant standards, impulse testing time may be up to 280 hours, which uses a lot of energy [12] to [15]. So it is of great importance to reduce energy consumption during impulse testing. The SAE proposed a standard for impulse machine equipment and operation, which has become the reference for most of later impulse testing equipment [16]. A new hydraulic intensifier has been proposed, which can be used for hydraulic impulse testing [17]. The ramp-output response of hydraulic impulse testing equipment and parameter adjustment for a standard pulse has been studied, and good performance of the system has been achieved [18]. All of these above mentioned systems can achieve the desired pressure curves. However, high energy consumption and energy waste are the main drawbacks. As the hydraulic accumulator can store and release energy over a short period of time, it is a good choice for energy regeneration in hydraulic impulse testing equipment, since the cycle time of hydraulic impulse testing is generally very short. In this paper, an energy regeneration system with a hydraulic accumulator is used for hydraulic impulse testing. The energy consumption and system efficiencies of the hydraulic impulse testing equipment with an accumulator are compared with those of hydraulic impulse testing equipment without an accumulator using mathematical analysis and simulations. The results show that the energy regeneration system can reduce energy consumption by 15 % and increase system efficiency from 62.82 to 75.71 %. Experiments have been carried out to study the control performance of the energy regeneration system for hydraulic impulse testing equipment with an accumulator, and the results show that a faster response speed and a larger pressure rise rate can be obtained. 1 SYSTEM DESCRIPTION A schematic diagram of the energy regeneration system with hydraulic accumulator for hydraulic impulse testing equipment is shown in Fig. 1. It mainly consists of the pressure source, servo valve, pressure-boost cylinder, accumulator, check valve, and test components. The pressure source, including a variable displacement pump, a check valve, a relief valve, and an accumulator, is used to provide hydraulic power for the system. Since some hydraulic components need to be tested under high pressure, a pressure-boost cylinder is utilized to achieve a pressure higher than that of the pressure source. The pressure-boost cylinder is designed to work in a differential mode, under the control of the three-way servo valve. The cylinder mainly consists of four chambers, in particular, the control chamber connected to the servo valve, system-pressure-chamber connected to the pressure source, the boost chamber connected to the test components, and the accumulator chamber connected to the accumulator. The test components can be hydraulic hoses, fittings, assemblies and so on. A check valve is used to supplement hydraulic oil for the test components. Fig. 1. Energy regeneration system for hydraulic impulse testing equipment It should be noted that there are two accumulators in Fig. 1 . One is connected to the accumulator chamber of the pressure-boost cylinder, the other is connected to the pump in the pressure source. Bladder accumulators are adopted due to their rapid response, long service life, and small dimensions. Both of them are helpful in reducing the energy consumption of the system. However, the accumulator connected to the pump is mainly used for absorbing pulsation of the pressure source, and is not the focus of this paper. Therefore, the accumulator referred to hereafter is the one connected to the accumulator chamber of the pressure-boost cylinder. For simplicity of analysis, the motion of the piston of the cylinder is divided into two strokes. When a positive signal is provided for the servo valve, the piston of the hydraulic cylinder moves to the right, so the hydraulic oil in the test components is compressed and correspondingly the pressure rises, which is defined as the boost stroke. In contrast, when a negative signal is provided for the servo valve, the piston of the hydraulic cylinder moves to the left, so the hydraulic oil in the test components is released and correspondingly the pressure decreases, which is defined as the return stroke. The key component of the energy regeneration system is the accumulator attached to the accumulator chamber of the pressure-boost cylinder. A variable damping is designed, in a mechanical structure way, between the accumulator chamber and the control chamber of the cylinder, as shown in Fig. 1. At the end stage of the boost stroke, the variable damping is so small that the accumulator may be considered as directly connected to the control chamber. Hydraulic oil can be provided by the pressure source through the servo valve. Then in the return stroke, the variable damping becomes larger. At the end of the return stroke, the damping is so large that the accumulator may be considered as isolated from the control chamber. Correspondingly, the gas in the accumulator is compressed, and the hydraulic energy in the test components and from the pressure source is recovered by the accumulator. In the next boost stroke, the energy stored in the accumulator can be released to help push the piston. To get better control performance of the output pressure, a feedback control method is adopted, with a pressure sensor detecting the pressure of the boost chamber of the cylinder, and a controller, which are neglected in the schematic diagram for simplicity. 2 MATHEMATICAL ANALYSIS To analyze the advantages of the energy regeneration system with hydraulic accumulator for hydraulic impulse testing equipment, comparisons should be done between hydraulic impulse testing equipment with and without accumulator, especially the energy consumption and system efficiencies. The hydraulic impulse testing system with accumulator is shown in Fig. 1, and the system without accumulator is the same as in Fig. 1, except that the accumulator connected to the cylinder and the corresponding chamber of the cylinder are omitted. For simplicity, as defined in Section 1, the boost stroke and return stroke may be analyzed separately under some conditions. 2.1 Power of the System without Accumulator 2.1.1 Load Power In the boost stroke, load power means the power consumption of the test components, and in the return stroke, load power means the output power of the test components. In both cases, load power of the hydraulic impulse testing system without accumulator is given as: Pl = PlIl , (1) where pL is the pressure of the test components, which can be measured with a pressure sensor, and qL is the flow rate at the inlet of the test components, which is given as: =■ dV_ dt (2) where VL is the volume of the test components, including the boost chamber of the cylinder connected together. According to the pressure formula of closed volume chambers: E • dVr (3) where E is the equivalent bulk modulus, determined by the bulk modulus of the test components Ec and the bulk modulus of the hydraulic oil Eo, and can be given as: 1 = 1 _L E " E+E • 2.1.2 Power of the Pressure Source (4) In the boost stroke, a positive signal is input to the servo valve, and the control chamber of the cylinder is connected to the pressure source through the servo valve. It should be noticed that the system-pressure-chamber of the cylinder is always connected to the pressure source. Therefore the pressure-boost cylinder works in a differential mode, and the power of the pressure source is determined by the system pressure ps, the difference of the input flow rate into the control chamber qx and the output flow rate from the system-pressure-chamber q2. For simplicity, it is assumed that the pressure drop of the servo valve is small enough to be omitted. Therefore the power of the pressure source in the boost stroke can be given as: PsB = Ps ■ -