Application of Artificial Neural Networks to Fault Detection of Rolling Ball Bearing

Abstract

Bearing is typical rotation parts in the machine. The failure of bearing under the machine operation can cause machine breakdown and also loss income. It is important for the industry to regularly check the condition of the machines for any malfunction because it can plan using efficient preventive maintenance. Nowadays, many factories use measuring devices to monitor and predict the machine’s conditions. When machine operates, vibration occurs and vibration pattern may be changed after machine operates for a while. In this study, the non-destructive testing by using vibration signal technique for bearing health monitoring was proposed in order to compare predicted results with various classification methods. In this experiment, bearings 6006z were simulated with six different conditions such as the normal condition of the bearing, the outer race fault, the inner race fault and the bearing fault with polishing grease of grit sizes of 100, 280 and 400, respectively. Vibration signals in time domain of each bearing type were detected by using accelerometer. Vibration signals were recorded into files. The vibration signals were interpreted using time domain and frequency domain techniques and compared precision of predicting bearing defect types using Vector Machine (SVM), K-Nearest Neighbor (KNN) and Artificial Neural Network (ANN). It is found that Artificial Neural Networks gave the best result of predicting bearing defect types with accuracy of 98%