Data-driven fault diagnosis method based on the conversion of erosion operation signals into images and convolutional neural network

Abstract

In the industrial process, the safety and reliability of the mechanical system determine the quality of the product, and whether small faults can be diagnosed in time is the key to ensuring the safe operation of the system and restraining the deterioration of faults. In recent years, the data-driven fault diagnosis has attracted widespread attention in academia. However, the traditional data-driven fault diagnosis methods rely on the features extracted from expert systems, so that the effect of fault diagnosis is entirely reliant on how well the expert system can extract the features. This paper proposes a new fault diagnosis method based on AlexNet Convolutional neural network (CNN) from a data-driven perspective. Firstly, a new method for converting time-domain vibration signal into RGB image based on erosion operation (EOSTI) is proposed. Initially converted three-dimensional (3-D) images have relatively close structural elements and are difficult to identify. For such defects, the target separated RGB image is generated. Secondly, explore the classification accuracy of AlexNet to make it more suitable for fault classification of different bearing datasets. Finally, the proposed method which is tested on two datasets, including coal washing machine dataset, maintenance fault dataset, has achieved prediction accuracy of 99.43 % and 99.67 %, respectively. The results have been compared with other methods. The comparisons show the effectiveness and accuracy of the proposed approach. The result shows that this method is feasible in engineering practice.