36 Journal on Processing and Energy in Agriculture 16 (2012) 1 Biblid: 1821-4487 (2012) 16; 1; p 36-40 Original Scientific Paper UDK: 338.43 Originalni naučni rad EXPERT SYSTEM FOR INDUCTION MOTOR FAULT DETECTION BASED ON VIBRATION ANALYSIS EKSPERTSKI SITEM ZA DETEKCIJU KVAROVA NA ASINHRONIM ELEKTROMOTORIMA ZASNOVAN NA ANALIZI VIBRACIJA Željko KANOVIĆ, Boris JAKOVLJEVIĆ, Milan RAPAIĆ, Zoran JELIČIĆ Faculty of Technical Sciences, 21000 Novi Sad, Trg Dositeja Obradovića 6, Serbia e-mail: kanovic@uns.ac.rs ABSTRACT This paper presents an expert system for induction motor fault detection based on vibration analysis and support vector machines (SVM). Vibration signals of healthy and faulty induction motors are collected and characteristic features, as indicator of fault pres- ence, are calculated, in both time and frequency domain. Two types of faults were considered, static eccentricity and bearing wear. Obtained feature sets were then used for training of support vector machines classifiers, a type of artificial intelligence classification technique which determines whether some of considered faults is present or not. An expert system for fault detection is designed com- bining a database of calculated features and trained SVM classifiers. This system was tested and validated on a number of healthy and faulty motors in the laboratory and in industrial facility for sunflower oil processing. Obtained results prove that this system can detect faults in early stages with high accuracy and reliability. Thus, it provides malfunction and failure prevention and improves overall performance and efficiency of industrial systems. Key words: fault detection, induction motor, vibration analysis, support vector machines. REZIME Praćenje stanja i dijagnoza kvarova na mašinama imaju važnu ulogu u sistemu održavanja, jer smanjuju troškove i poboljšavaju produktivnost, efikasnost i iskorišćenje mašina. U ovom radu predstavljen je ekspertski sistem za detekciju kvarova na asinhronim elektromotorima baziran na analizi vibracija i potpornim vektorima (SVM). Analiza vibracija primenjena je zbog svoje visoke tačnos- ti i pouzdanosti. Snimljeni su signali vibracija više tipova ispravnih i neispravnih elektromotora, pomoću kojih su izračunata karakte- ristična obeležja, koja predstavljaju indikatore prisustva pojedinih kvarova. Razmatrana su dva tipa kvarova, statički ekscentricitet rotora i oštećenje ležajeva. Karakteristična obeležja su primenjena za obuku SVM klasifikatora, baziranih na veštačkoj inteligenciji, koji detektuju prisustvo kvara. Kombinovanjem obučenih SVM klasifikatora i baze podataka sa snimljenim signalima, napravljen je ekspertski sistem za detekciju kvarova, koji je ispitan u laboratorijskim uslovima i u postrojenju za preradu suncokretovog ulja. Dobi- jeni rezultati pokazuju da ovaj sistem sa visokom tačnošću i pouzdanošću može detektovati kvarove u ranim stadijumima, te da stoga omogućava prevenciju kvarova i otkaza i poboljšava performance i efikasnost industrijskog sistema. Ključne reči: detekcija kvarova, asinhroni elektromotor, analiza vibracija, klasifikatori sa potpornim vektorima. INTRODUCTION Induction motors play an important role as prime movers in manufacturing, process industry and transportation due to their reliability and simplicity in construction. Although induction motors are reliable, the possibility of unexpected faults is un- avoidable. The issue of robustness and reliability is very impor- tant to guarantee the good operational condition. Therefore, con- dition monitoring of induction motors has received considerable attention in recent years. Early fault diagnosis and condition monitoring can reduce the consequential damage, breakdown maintenance and reduce the spare parts of inventories (Matić et al., 2010). Moreover it can prolong the machine life and increase the performance and the availability of the machine. Many re- searchers have proposed techniques and systems for doing the diagnosis process. Various techniques have been used, such as motor current signature analysis (Kulić et al., 2010), electro- magnetic torque measurement (Thollon et al., 1993), acoustic analysis (Lee et al., 1994) and partial discharge (Stone et al., 1996). However, the most popular techniques are vibration analysis and stator current analysis due to their easy measurabil- ity, high accuracy and reliability. Support vector machines (SVMs) have been extensively employed to solve classification problems. In machine condition monitoring and fault diagnosis, some researchers have used SVMs as a tool for classification of different kind of faults, such as ball bearing faults (Jack and Nandi, 2002), gear faults (Samanta, 2004), condition classifica- tion of small reciprocating compressor (Yang et al., 2005a), cavitation detection of butterfly valve (Yang et al., 2005b) and so on. To perform good classification using SVMs, the preparation of data inputs for classifier needs special treatment to guarantee the good performance. Recently, the use of feature extraction and feature selection for data preparation to avoid the redun- dancy before inserting into classifier has received considerable attention (Cao et al., 2003). There are numerous papers and studies that describe laboratory experiments and applied tech- niques conducted in purpose of different kinds of induction mo- tor faults detection. However, application of these techniques and their results in real industrial systems is not so common. In this paper, an expert system for induction motor fault detection is presented, which represents an attempt to implement very well known fault detection techniques in real industrial system. This system consists of several modules, which perform all necessary tasks in fault detection and classification process (vibration sig- nal acquisition, data processing, SVM fault detection and visu- alization and archiving of obtained results). Two kinds of faults are considered, bearing wear and static eccentricity. The system is developed and tested using real data from laboratory and sun- flower oil processing industry. It can be used as stand-alone tool for condition monitoring of induction motors. Use of this system in industrial facilities could assist in early fault detection and prevent malfunctions and failures of production systems, im-