Classification Method for Industrial Fault Images Based on Hyperparameter Optimization of MobileNet Using Progressive Mesh Method

With the rapid development of artificial intelligence and image recognition technology, the demand for efficient and accurate fault detection methods in the industrial sector has been steadily increasing. Traditional methods often rely on human experience and existing knowledge, which may result in insufficient accuracy and efficiency. This study aims to propose a classification method for industrial fault images using a MobileNet optimized by the Progressive Mesh Method, thereby alleviating the burden on workers and improving the accuracy of fault detection. One-dimensional signals collected are transformed into image form using a two-dimensional wavelet transform. For the image recognition problem, a convolutional neural network (CNN) architecture is employed. Among lightweight CNN architectures with less than 10M parameters, the MobileNet network exhibits the best noise resistance performance through noise resistance testing. To fully exploit the advantages and practicality of the MobileNet network, the Progressive Mesh Method is used to optimize the learning rate and the number of minimum batch samples. The pre-trained MobileNet network is fine-tuned to meet task requirements, and prepared training samples are used for network training. The performance of the network model is validated using the dataset from the Case Western Reserve University Bearing Data Center. Experimental results demonstrate that the robustness and noise resistance of the entire network framework have been enhanced through noise resistance testing and the Progressive Mesh Method, proving that the proposed method has certain applicability and promising potential for industrial fault image classification applications.