International Conference on AdvancesandInnovations in Engineering (ICAIE) 330 Classification of Bearing Fault Size by Using Support Vector Machines Kaplan Kaplan 1 , Melih Kuncan 2 , H. Metin Ertunç 3 1, 2, 3 Department of Mechatronics Engineering Kocaeli University, Izmit / Kocaeli kaplan.kaplan@kocaeli.edu.tr melih.kuncan@kocaeli.edu.tr hmertunc@kocaeli.edu.tr Abstract — Bearings are generally used as rolling elements in rotation machines. Faults in the rolling elements causes breakdown, and this may lead downtime and huge damages in rotating machines. On the other hand, bearings are often employed under high load and high running speed conditions. In this study, artificial faults are created on bearing inner rings by a laser beam in certain size namely 0.15 cm, 0.5 cm, 0.9 cm diameter. Vibration signals are collected by a data acquisition device in a shaft-bearing test setup. Before classifying the data, feature extraction is performed to characterize the signal. Statistical features are calculated and they are used as input to classification method. SVM classification model is employed to diagnose the size of the faults. The SVM model developed in this study classify the size of bearings faults with no prediction error. In addition, 0.1 mm error band is determined to eliminate minor bugs. Keywords- Support vector machines; bearings; diagnosing I. INTRODUCTION A bearing element is composed of an inner ring, an outer ring and certain number of rolling balls. Bearings are also one of the most important elements of mechanisms such as motor-shaft which is rotating with the least friction. Nowadays, there are usually bearings on the wheel and on the hub of everything that turns. The first manufacturer of bearings is the FAG Company and its founder Friedrich Fisher invented the first bearing in 1883. There are nearly three hundred thousand types of roller bearings used in the industrial applications. Hence, the usage areas of the bearings are a lot. They are used in many different areas and sectors as a transitional element between mechanisms such as aircrafts, subways, coaches, buses, trains, engines, conveyor lines, gears, patters, washing machines, microscopes, telescopes, wind power plants, pumps [1]. A fault on the roller bearings may cause damage in the whole mechanical system. Moreover, faulty bearings can lead to downtime in whole plant. This is undesirable case especially for the big and expensive systems. For this reason, the bearings must be checked at periodic intervals; and repair or replacement must be carried out before fault occurs. To prevent these faults, many diagnostic studies carried out. There are numerous research studies in the literature related to diagnosing of bearing elements. For example, Liu et al. [2] developed an expert system that can detect faulty conditions by examining the vibration signals measured in the radial direction before bearing faults occur. This system can classify bearing faults by using fuzzy logic model with various time domain features such as kurtosis. The authors also calculated the success of the diagnosis methods and they claimed that the fuzzy logic method can detect fault with 100% success. The developments in fault diagnosis by using artificial intelligence strategies are summarized by Frank and Koppen –Seliger [3]. They pointed out that the artificial intelligence methods play an important role in the control of faults together with analytical methods. Along with the developing classification methods, Abdulshahed et al. [4] employed ANN and ANFIS model to detect warming points by friction, corrosion etc. They recorded the thermal warming areas that occurred during spindle motor rotation on the CNC machine tools. The correctness of the positioning of these thermal faults that is very important in terms of machine stability. ANN and improved c-averages fuzzy method achieved valid classification rate. Under different operating conditions, the ANFIS model carried out 98.6% successful classification rate. Zhang X. et al. [5] performed fault detection and classification of motor bearing by using a hybrid model. The authors calculated permutation entropy (PE) from the vibration signal. They employed the support vector machines (SVM) optimized by inter-cluster distance (ICD) in the feature space in order to classify the fault type and fault severity. They noticed that the developed model obtained effective and robust detection and classification results for motor bearing fault. Fernandez-Francos D. et al. [6] studied on an automatic method for bearing fault detection and diagnosis. They employed a one-class v-SVM to classify normal and faulty conditions. The authors used band-pass filters and Hilbert Transform to obtain the envelope spectrum of the original raw signal. To compare the performance of the method, they usedtwo different data sets are used that are real data from a laboratory test setup and a fault-seeded bearing test setup. They claimed they determined not only the failure in an incipient stage but also the location of the defect.