Convolutional neural network (CNN) for fault diagnostics of rolling element bearing

Convolutional neural network (CNN) for fault diagnostics of rolling element bearing

This paper discusses the application of conventional machine learning (ML) methods in detecting bearing faults. The recent advancements in deep learning algorithms have renewed interest in intelligent machine health monitoring in both industry and academia. A comparative study is conducted on the classification accuracy of the Convolutional Neural Network (CNN) algorithm using the bearing dataset resources in Prognostics and Health Management (PHM (to provide a more intuitive insight. The PHM data representing 2 different loads are considered: First operating conditions: 1800 Revolutions Per Minute (rpm) and 4000 Newton (N), Second operating conditions: 1650 (rpm) and 4200 (N). The dataset of the first condition is used to train and test the CNN model, while the second dataset is employed to test the CNN model with different conditions. Then, the accuracy of the network is compared in these two situations. The results show that the accuracy of the CNN model is 96% for training and 93% for testing data. The CNN model accuracy for the other conditions is 85%. As a suggestion, Fast Fourier Transform (FFT), Envelope, and Transfer Learning can be utilized to improve the results.

This paper discusses the application of conventional machine learning (ML) methods in detecting bearing faults. The recent advancements in deep learning algorithms have renewed interest in intelligent machine health monitoring in both industry and academia. A comparative study is conducted on the classification accuracy of the Convolutional Neural Network (CNN) algorithm using the bearing dataset resources in Prognostics and Health Management (PHM (to provide a more intuitive insight. The PHM data representing 2 different loads are considered: First operating conditions: 1800 Revolutions Per Minute (rpm) and 4000 Newton (N), Second operating conditions: 1650 (rpm) and 4200 (N). The dataset of the first condition is used to train and test the CNN model, while the second dataset is employed to test the CNN model with different conditions. Then, the accuracy of the network is compared in these two situations. The results show that the accuracy of the CNN model is 96% for training and 93% for testing data. The CNN model accuracy for the other conditions is 85%. As a suggestion, Fast Fourier Transform (FFT), Envelope, and Transfer Learning can be utilized to improve the results.