A new deep learning model combining CNN for engine fault diagnosis

Abstract

Real-time condition monitoring of electric motors and early diagnosis is of great importance for ensuring safe and reliable operation, preventing major accidents, and reducing production costs. Therefore, many intelligent fault diagnosis methods have been proposed. However, in industrial applications, the constantly changing loads of electric motors and the inevitable noise from the working environment cause a decrease in the performance of intelligent fault diagnosis methods. In this study, an effective and reliable deep learning model named the Combined One and Two-Dimensional Deep Convolutional Neural Network with Wide First-layer Kernels (WDD-CNN) is proposed for real-time condition monitoring and early fault diagnosis under noisy and changing operating conditions. The primary contribution of this study is the development of a fault diagnosis method that can operate in real-time to provide early detection of faults that may occur in electrical drive systems under operating conditions that are unpredictable and noisy. In addition, the proposed model works directly on raw signals, eliminating the complexity of preprocessing processes. The Case Western Reserve University (CWRU) dataset is used to test the performance and effectiveness of the proposed WDD-CNN model under different load conditions and for noise suppression. Additionally, the effectiveness of the model against data coming from a single sensor channel is also tested, and the results are recorded. The proposed method achieves 100% accuracy when tested with normal signals. Comparative results reveal that the WDD-CNN model outperforms other current state-of-the-art methods with an accuracy rate of 96.45% under different operating loads. © 2023, The Author(s), under exclusive licence to The Brazilian Society of Mechanical Sciences and Engineering.