J Intell Manuf (2009) 20:243–247 DOI 10.


An experimental study, about detection of bearing defects in inverter fed small induction motors by Concordia transform
˙ ˙ Izzet Yilmaz Önel · Engin Ayçiçek · Ibrahim Senol ¸

Received: 1 April 2006 / Accepted: 1 December 2007 / Published online: 4 January 2009 © Springer Science+Business Media, LLC 2009

Abstract This paper describes an application about detection of bearing defects in inverter fed induction motors, using Concordia transform approach based algorithm. After introduction, brief information is given about bearing structure and type of bearing failures. Next section, Concordia transform theory is mentioned then, RBF neural network structure is summarized. After that, test system information is specified. This paper indicates that Concordia transform approach is a reliable tool to detect bearing faults in inverter fed small induction motors. The generality of the proposed methodology has been experimentally tested on a 1 HP squirrel-cage induction motor. At the end of the paper, an ANN algorithm is proposed that could detect the bearing faults automatically. The obtained results have 93.75% accuracy. This study suggests that proposed Concordia transform based fault detection algorithm could be integrated in an induction motor driver so, bearing condition of the induction motor could be observed while motor is working and bearing faults could be detect before they become serious. Keywords Induction motor · Bearing faults · Concordia transform · RBF neural network

Introduction Induction motors are used widely in industrial applications due to their low cost and no care needed structure. Although their robust design, some faults could be occur in induction motors. In the reliability and performance of all motor systems, bearings play an important role. Due to the close relationship between motor system development and bearing assembly performance, it’s difficult to imagine the progress of modern rotating machinery without consideration of the wide application of bearings. In addition, the faults arising in motors are often linked with bearing faults. The researches show that bearing problems account for over 40% of all rotor failures (Önel et al. 2005). That’s why; detecting the bearing faults in induction motor has significant importance. There are many studies about detecting bearing faults in induction motors (Schoen et al. 1995; Benbouzid et al. 1999; Zidani et al. 2003). Most of the studies utilized motor current signature analysis (MCSA) that use Fast Fourier Transform. When the induction motor is driven with an inverter, the levels of harmonic frequencies are increased and it becomes more difficult to detect bearing faults using FFT algorithm. This paper proposed a Concordia Transform based algorithm to detect bearing faults in inverter fed induction motors.

Bearing failure
˙ Önel (B) · E. Ayçiçek · ˙ Senol I. ¸ I.Y. Electrical-Electronics Faculty, Electrical Engineering Department, Yildiz Technical University, 34349 Besiktas-Istanbul, Turkey e-mail: ionel@yildiz.edu.tr E. Ayçiçek e-mail: eaycicek@yildiz.edu.tr I. Senol ¸ e-mail: senol@yildiz.edu.tr

A bearing consists of two rings inner and outer, between which a set of balls or rollers rotate in raceways. Figure 1a shows the parts of a deep groove ball bearing. There are many reasons to make bearings faulty. Dust and corrosion are the main factors. Induction motors are operated in hard conditions like factories. Dust and other foreign materials could contaminate easily the grease and damage


This causes raise in frictional losses and excessive heating. the mains current has no homopolar component. EMRC 2007). the connection to the mains does not usually use the neutral. sin ωt − (4) 2 2 where I . If dissimilar grease is used. so faults could be detected automatically with this method (Nejjari and Benbouzid 2000). Therefore. After discharge. Installation problems are often caused by improperly forcing the bearing onto the shaft or in the housing. the voltage again accumulates on the shaft and the cycle repeats itself. i b . a suitable one being based on the current Concordia vector (Benbouzid 2003). which induces a voltage on the rotor shaft. Concordia vector theory In three phase induction motors. As a function of mains phase variables (i a . time variable. It should be used special tools like bearing heaters to install the bearings to prevent this failure. b bearing current path way (Reliance Electric Company 2000) id = 2 1 1 ia − √ ib − √ ic 3 6 6 (1) the oil film between ball and raceways. deteriorated lubrication and even perspiration from careless handling during installations (Obaid et al. Concordia transform describes D and Q axes which are stationary according to the stator. and the first layer connections 123 . Once the chemical reaction has advanced sufficiently. In induction motor drivers. 1958). 1b shows the path of bearing current (Greenheck Fan corporation website 2007. 2. then 1 1 iq = − √ ib − √ ic 2 2 (2) Under ideal conditions. the switching frequency of semiconductor components ranges from 1 kHz up to 20 kHz and is referred to as the carrier frequency. Induction motor stator current can be defined as a vector which is sum of D and Q components. √ Radial basis function networks A schematic of the RBF network with N inputs and a scalar output is given in Fig. maximum value of the supply phase current. This voltage referred to as common mode voltage or shaft voltage. This is very simple reference figure that allows the detection of abnormal conditions by monitoring the deviations of acquired patterns. This random and frequent discharging cause pitting of the bearing’s rolling elements and raceways Fig. t. In addition this pattern could be learnt by an artificial neural network. particles are worn off resulting in the same abrasive action produced by bearing contamination. supply frequency. ωs . 1958). 1 a The structure of a deep groove ball bearing. The input nodes pass the input values to the connecting arcs. i c ) the current Concordia vector components (i d . sin ωt (3) id = 2 iq = π 6 I. 2003. a chemical reaction could be occur and this reduces the effect of the lubrication. Another cause of bearing failure is bearing current. Improper lubrication is another bearing failure factor and it contains under. acids.244 J Intell Manuf (2009) 20:243–247 the voltage discharges in a short pulse to ground through the bearing. Bearing corrosion is produced by the presence of water. The basic structure of the network could be seen with this figure. over lubrication and using different grease inside the bearing. Due to the ratio of change of the V / T creates a parasitic capacitance between the motor stator and the rotor. Eschmann et al. Its representation is a circular pattern centered at the origin of the coordinates. This produces physical damage in the form of brinelling or false brinelling of the raceways which leads to premature failure (Eschmann et al. three phase currents lead to a Concordia vector with the following components: √ 6 I. i q ) are Fig. Shaft voltage accumulates on the rotor until it exceeds the dielectric capacity of the motor bearing lubricant. A two dimensional representation can then be used for describing three phase induction motor phenomena.

x. In this study. 4 Four defected bearings that were used in experimental studies. The motor was loaded with car alternator VDN10701118 that 12 V DC output is applied to the variable resistive loads. d ball defected ball bearings 123 .75 kW two pole induction motor that parameters are given in Table 1 below. The LabVIEW programme converts the sampled signal that sampling frequency is 10 ksample/s. a Outer race defected.σ) y Fig. The currents that flowing in three phase of the induction motor. Electrical part of the test jig has three parts: three current transformers. weighted actions in second layer only. 3 Test jig has two parts: mechanical and electrical. System information The purpose of the monitoring system is to measure threephase the induction motor stator current. The radial basis function network does not extend to more layers in as natural manner as backpropagation networks because of the different functions and parameters of the layers.95 A 2.75 kW 50 Hz 220/380 V 3. MATLAB programme is used to obtain D and Q components of the motor current utiliz- ing Concordia transformation.780 rpm 1 Fig. The nominal current is IN = 1. c cage defected. The second layer of connections is weighted. b inner race defected. tacho generator and the induction motor were connected together with steel couplin in the same alignment. Mechanical part consist a tacho generator for speed measuring and an induction motor that has bearing failure. but improved training and performance are directly related to this architecture (Leonard and Kramer 1991). Thus each hidden nodes receives each input value.4/1. The hidden nodes are the radial basis function units. a car alternator to load the induction motor. The transfer function for hidden nodes is nonmonotonic in contrast to the monotonic sigmoid function of backpropagation networks. and the output nodes are simple summations. unaltered. 2 A radial basis function network architecture with nonmonotonic transfer function in hidden nodes. Fig.95 A when star connected to 380 V network. and summation units for output nodes Table 1 Rated parameters of the machine under test Power Frequency Voltage ( /Y ) Current ( /Y ) Speed Pole pair (p) 0. are sensed by a current transformer and they are sent to the National Instruments PCI-1200 data acquisition card that is connected to the PCI slot of the personal computer (Fig. 5). an outlet to the PC based data acquisition card and bulbs that are used as load are not weighted.J Intell Manuf (2009) 20:243–247 Radial Basis Units Inputs Outputs 245 x W ^ a(x. was used. 0. The alternator.

deep groove ball bearings.2ZR type eight bearings were defected with Fig. ABB ACS140 type induction motor driver is used in experiments. type 6204.000 rpm. The driver has 1 to 3-phases. PC Fig. Testing equipment is illustrated in Fig. The V/Hz control mode is preferred.246 3 phase Induction motor J Intell Manuf (2009) 20:243–247 R S LOAD Tacho Generator T Data Acquisition Card Low Pass Filter Park’s Transform The bearings of the induction motor are single row.2ZR. cage and ball defect) Fig. In the Electrical Machines Laboratory of Yıldız Technical University. but 50 Hz 380 V AC output is used to feed the motor. 6 Comparison of the induction motor stator current D–Q components with healthy and different type faulty bearings 123 . ±10% permitted tolerance input and 3 phases 0–300 Hz output. 4. 5 Measurement configuration block diagram Experimental study The analog tacho generates the 90 V in 3. Switching frequency is chosen 16 kHz. 200–480 V. 6204. 3. Bearings have eight balls. Experiments were applied to nine bearings: one of this is healthy and eight of bearings were defected (inner race. outer race.

(1991). Reliance Electric Company. (2003). Nejjari. et al.. Dalcı. While these were not a realistic bearing failure.1} LW{2.. E. cage and ball. H. So. Sampled data was applied to Concordia transform to obtain D and Q axis components. 1274–1279. 123 . Induction motor three phase current were sampled and transferred to PC by data acquisition card. References Benbouzid. Heyden. 14(1). B... inner race. Detection of outer raceI.electricmotors. and application. Schoen Randy. A RBF neural network is designed to automate the fault detection process. 18(4). EMRC (Electric Motors Reference Center) web site.. Induced bearing currents. IEEE Transactions on Industrial Applications. The sampling frequency of the motor current is 10 ksample/s. Induction motor’ fault detection and localisation using stator current advanced signal processing techniques. 11(3). There are two outputs of the ANN. USA. design. Önel. Leonard. M. G. et al. H. Ball and roller bearings: Their theory. IEEE Control Systems Magazine. Power Electronics and Drives. Motor bearing damage detection using stator current monitoring. 713–722. Figure 6 shows the D and Q components’ variation of healthy and faulty motor.reliance. way bearing defects in small induction motors using stator current analysis. (2003). 14–22. (2007). After Concordia transformation 200 D and 200 Q components of the motor current are obtained. Monitoring and diagnosis of induction motors electrical faults using a current Park’s vector pattern learning approach. London: K. H.. M. Atlanta. 730–735. 2003. (2007). ˙ Y. outer race.. Application guides. In SDEMPED 2003.greenheck. This study suggests that proposed Concordia transform based fault detection algorithm could be integrated in an induction motor driver so. H. P. A. 30(Part 6). August 24–26.1} O1 I 400 400 b{1} b{2} O2 2 50 Fig. J. K. F. R. & Kramer. Induction motor stator faults diagnosis by a current Concordia pattern based fuzzy decision system.com. IEEE Transactions on Energy Conversion. Zidani.J Intell Manuf (2009) 20:243–247 247 I1 IW{1.. http://www. 31(6). http:// www. The experimental results show that Concordia transform based algorithm can be used to detect bearing defects in inverter fed small induction motors. Radial basis function networks for classifying process faults. IEEE Transactions on Power Electronics.com/. et al. If O1 is active means bearing has some fault. It can be seen that the faulty and healthy bearings could be detect with examining D–Q components of the motor current. GA. Symposium on Diagnostics for Electric Machines. (2000). & Benbouzid. Eschmann. 31–38. 18(4). 3×200 = 600 samples are taken from three phases of motor current in (20 ms) a period. et al. 469–475. ˙ (2005). bearing condition of the induction motor could be observed while motor is working and bearing faults could be detect before they become serious. Induction motor stator faults diagnosis by a current Concordia pattern-based fuzzy decision system. Results and suggestions This paper has investigated the feasibility of detecting bearing faults using a Concordia transform based algorithm in inverter fed three phase induction motors. E. the artificial bearing fault produced similar effect on stator current. M. Sadhana. et al. (2000). CurrentShieldTM technology— motors and drives for cleanroom application. (2003). 36(3). 7 The RBF neural network structure that is used to detect bearing faults A radial basis function neural network was designed to automate the fault detection process and it succeeds with 93.. E. R. ¸ I.75% accuracy to determine the bearing faults in inverter fed induction motor. M.. (1958). If O2 is active means bearing has no fault. A. IEEE Transactions on Industry Applications. These components are the inputs of the BRF neural network. (1995). Benbouzid. Obaid.machinedesign. IEEE Transactions on Energy Conversion. Greenheck Fan corporation website. com. 469–475. 7. & Senol. et al. Stator current analysis for bearing damage detection in induction motors. http://www. (1999). The RBF neural network structure is given in Fig.