FAULT DIAGNOSIS OF HIGH SPEED ROLLING ELEMENT BEARING USING SOFT COMPUTING TECHNIQUES

Abstract

Majority of products or machines in the world are of rotating nature, bearings are the very basic and utmost important component required for rotation. Various big machinery, precious, costly components rely over bearing for its desired motion. Although the cost of replacing bearing with new one is not that much but as mentioned, various other components relying over it are costly and precious. In order to save this other component from failure, we may have to avoid sudden failure of the bearings over which they. rely. For the same reason we need to have proper knowledge about when and how bearing fails, which can be done by proper diagnosis of rolling element bearings. In recent times various soft computing techniques have been used for fault diagnosis. So we have chosen fault diagnosis of high speed rolling element bearing to be done by using soft computing technique. For fault diagnosis we have used experimental setup which contains shafts fitted with five different kind of bearings out of which one is healthy bearing and rest others are faulty with fault in inner race, outer race, spall over ball and combined all three defect in one. This shaft successively mounted over the motor and allowed to run. Vibration signatures obtained from the bearing due to fault in it is captured by using CoCo-80 machine, from this raw vibration signature various features are extracted. Total 16 features are extracted out of which 8 are found to be efficient including both horizontal and vertical features. By using this effective extracted features and different classifiers with the help of WEKA software, we have done soft computing for fault classification and compared the classification efficiency of artificial neural network, support vector method and APF-KNN. Although all the three classifier have given good result but it has been found out that the fault classification efficiency of APF-KNN is higher than other two.