One-Dimensional Convolutional Neural Network (1d-Cnn) for Bearing Element Fault Diagnosis

Mechanical machines are considered as significant progress in the area of large-scale, high-speed, accuracy, systematization, and computerization. Rolling bearings are the key component of any mechanical equipment. Therefore, a health monitoring system is needed to guarantee the safety of the machine. Rolling element bearing model depends upon the vibration signal. This vibration signal is preprocessed, features are extracted and then input into the classifier. This classifier is used to classify the faults. This type of fault diagnosis method relies on the human experience selection in feature extraction. As a result, an incomplete and inadequacy of feature extraction are obtained. It also requires more time in designing features. Therefore, the human experience-based feature extraction method finds difficulty in guarantee versatility. This work is established to study the important part bearing in the machine. One-dimensional convolutional neural network (1D-CNN) is proposed with two convolutional layers to diagnose time-domain vibration signals. The model's recognition rate from Case western reverse university (CWRU) bearing database has achieved more than 98% accuracy by using data set enhancement techniques. 1D-CNN directly acts on the time domain vibration signal to diagnose the bearing faults. A unified framework for convolutional neural network algorithms has been employed for bearing fault diagnosis namely wide depth convolutional neural network (WDCNN) model whose structure emphasizes the characteristics of a large kernel for the first layer. The WDCNN model has resulted in a more than 99% recognition rate.