Informatica 35 (2011) 211-219 211 Fault Diagnostics of Centrifuge Pump Using Data Analysis in Spectrometric Method Mansour Esmaeilpour and Elnaz Nomigolzar Engineering Dept., Islamic Azad University-Hamedan Branch, Hamedan, Iran E-mail: ma.esmaeilpour@gmail.com Mohamad Reza Feyzi Derakhshi Engineering Dept., University of Tabriz, Tabriz, Iran Zarina Shukur Faculty of Information Science and Technology, National University of Malaysia, Malaysia E-mail: zs@ftsm.ukm.my Keywords: vibration analysis, spectrometry, centrifuge pomp, kurtosis and fault diagnostics Received: October 1, 2009 Vibrations analysis is one of the main surveying methods in maintenance and fault detection of machines in the industry. This method has unique advantages and disadvantages relating to surveying and fault detection of the machine. Objective of this research is to show the relationship between vibrations analysis and fault detection. The major problem of the vibration analysis is using the sensitive aural of the vibration sensors by human experts. On the other hand, human fault is time consuming which shows the position of the proposed method that by removing the human factor and increasing the speed and accuracy of the fault diagnosis course to increase the performance of the proposed method. Faults detection of the equipment is one of the most suitable ways of caring about the device when the equipment is on. Predictive repairing methods are new types of preventive repairs which use modern measurement and processing techniques for accurate fault finding and accessing technical conditions of the devices during exploitation and specification when maintenance and repairing operation are needed. In order to study the vibrations in pumps and to find its fault, different conditions of pump (sound and defective) were investigated. By putting sensors in a horizontal and vertical direction, the information was recorded and data vector activity was done for several times. Data gathering process was performed on the sample of centrifuge pump for its fault diagnostics through spectrometry. The results of analysis were able to distinguish between sound and defective data by studying on its acceleration range but there were analyzed and studied for ensuring data of different rounds which was selected randomly. This method is compared with Multilayer Perception Artificial Neural Network form terms of the processing time and accuracy that the result shows the superiority of the proposed method. Time data spectrum which had range 8000 N/mm and higher was sound and in interval between 1000 N/mm and 2000 N/mm, there was a need to repair and in interval 1000 N/mm and lower, it was defective. Povzetek: Predstavljena je metoda diagnosticiranja črpalk na osnovi vibracij. 1 Introduction Some of the phenomena which occur in the nature are specified well with statistical means and for this reason; one should have statistical look at random changes in such signals [7]. For example, self-correlation function of random process in time domain and its Fourier Conversion which are called power density spectrum are of the applied tools for analyzing these signals. On the other hand, there is an important class of signals such as static random process which doesn't have limited energy and as a result Fourier Conversion. Such signals have limited average power and are specified with power density spectrum. For machinery pump fault diagnostics, centrifuged pumps in repair and maintenance section of Idem Company have been selected. For studying faults of machinery pumps which are one of the integral parts of installations of each plant, firstly it is necessary to specify the importance of this part of plant and secondly how sensitive the Fault diagnostics of such parts is [1]. For this reason, we will refer to predictive net systems which have been commissioned in most plants and such systems removes need for studying fault diagnostics in machinery equipment such as centrifuge pumps. 260 Informatica 35 (2011) 259-268 M. Esmaeipour et al. 2 Maintenance and repair methods (MaR) Since early 1970s, changes in the industry have led to important movements in MaR. Changes which led to industrial mutation can be classified into three fields [3], 1. The formation of new requirements 2. New research 3. New MaR techniques Nowadays, one can show the applied changes by commissioning the following 2.1 Methods in plants Among mentioned methods, only part 3 is discussed which is stronger in application. American Production and Inventory Control Society is kind of preventive repair and maintenance on the basis of definition which denies MaR activities [2], This MaR uses unnecessary statistical analyses (and even destructive) and predicts the time which preventive MaR activities should be executed on devices and equipment. Detection of faults of the equipment is one of the most suitable ways of caring about the device when the equipment is on. Predictive repair methods is new type of preventive repairs which use modern measurement and processing techniques for accurate fault finding and getting access to technical conditions of the devices during exploitation and specification when MaR operation is needed [6], Predictive MaR techniques such as vibrations analysis, infrared heat images, X ray, ultrasound fault diagnostics and something like these have allowed predicting the conditions of the device and decision about necessaiy repairs [4], One can specify the position of Fault diagnostics using vibrations analysis and other methods in terms of application in the plants in case of commissioning such systems with suitable efficiency. After this stage, it is necessary to perform a data gathering phase for recording vibrations from the sample of the selected pumps. With regarding the applicability of the plan, the samples were selected with special criteria [5], For this purpose, one sample of the pump out of the sound pumps, one sample of the pump out of the defective pumps and one sample of the pump out of the sound pumps with long term function which have been used due to lack of defects are selected. In averaging method, we consider time pulse of a round (on the basis of the shortest time of a round) on the basis of the shortest time by using this frequency and the obtained signals for the rounds with the longest time are renovated on the basis of the shortest time. In this way, the problem of the asynchrony of the signals is solved to some extent. Then time averaging is done on these signals [10], In this method, accelerometer has been put on the pump and the outlet signal is recorded from it in a time interval equivalent to time of a round of circulation of the pump. Another example in the next time interval is averaged. When this process was done with the use of frequency sampling, the effect of the parameters relating to tested pump such as strokes resulting from collision of the vane in average signal increased and the specifications relating to other pumps and other factors such as noise were excluded from the average. For distinguishing between the intervals, a tachometer is used for the creation of the pulses or frequencies equivalent to reference shaft rotation frequency. Then, this pulse frequency is changed on the basis of ratio of the speed which is obtained according to the number of vanes of the engaged pump in such a manner that the limited length of the sampling is considered to be equivalently to a round of pump. The results are recorded as a diagram of average signals in which the number of signal equals to the number of rounds of the pump. 2.2 Data spectrometry Analysis techniques in the area of frequency for the vibration of signals and their related processing as well as results related to spectrum analysis give us clear information about rotary sets fault detection [8], Therefore, by using of this method, one can separate important faults from each other and identify them: Imbalance Failure of vans Failures of ball bearing resonance Studying on the performed research shows that ball bearing clearance decreases the range of vibration in the frequency and will increase its harmony [7], In this paper, peer to peer frequency results of pump signals are studied with the reference signals which have been used. Regarding the fact that it is possible to change the speed of the pump from revision to revision, in signal time field averaging method, peer to peer study will not be possible. As a result, for doing such direct comparison between signals, Fourier Conversion of the time field average signals is calculated. This method and other methods which are common in industry may detect only the fault and do not give enough information about the place and degree of fault [9], More importantly, computer software's and systems for faultfinding the simple machines such as pumps; ball bearing and etc. have been designed more widely and used in industry but research continues on some more complex systems faultfinding such as pumps. 3 Machinery pump and a possibility to provide data gathering medium Data gathering system is pulse type and sensor used for accelerating the body of the pump which has been made of Danish Company with 4 channels of B&K, type 4508 and is installed in ball bearings which are FAULT DIAGNOSTICS OF CONFIGURE PUMP Informatica 35 (2011 ) 259-268 261 As shown in Figure 2, two sections of the photo have been shown with a white color and on the right side is an acceleration sensor and on the left side is a label sensitive to the laser light which indicates rotation of the tachometer. In Figure 3, pulse device, tachometer, a computer the laser tachometer and the manner in which b they are positioned are shown. The software which is recording data is completely specified. In Figure 4, the position of tachometer (opposite to label sensitive to laser) is shown clearly. Figure 4: Tachometer positioning manner for recording output shaft speed After performance of data gathering phase, it is necessary to take an action regarding the reduction of noise on it and data averaging is one of the methods which are used as follows: 3.1 Stages of doing research on samples of machinery pump 1. Performance of a data gathering phase for samples of machinery pump 2. Studying data and if necessary the performance of noises reduction methods. After performing data gathering stage, data was prepared for doing different studies. In each stage data gathering was done 100 times. On the other hand, it entered spectrometry stage. We can see some samples of raw data with different rounds in the flowing figures. Therefore, raw data entered in data analysis program. two channels for tachometer and two channels for acceleration. Tachometer is also the type 0023 made of the same company. Totally 20 seconds were considered for each one of them and the analysis was done in terms of time. This was done by the cooperation of Idem Plant and University of Tabriz and Faculty of Mechanics by conclusion of a contract. Before data gathering, we need to signal synchronous by averaging. In Figure 1 we can see the Averaging process of two signals. Figure 1 : Averaging process of two signals. Figure 2: The position of the sensor on output shaft of centrifuge pump (right side) and an indicator which is sensitive to a laser (left side). INCREASED SCALE —W- Frequency spectrum transformed back into time domain to give enhanced signal average Frequency spectrum of signal avefage determined and non-defect related frequancies suppressed MESH FUNDAMENTAL MESH 2nd HARMONIC Signal average showing vibration information Figure 3: Data gathering manner with the related software and pulse device 262 Informatica 35 (2011) 259-268 M. Esmaeipour et al. 150 100 est S 50 5«