A novel fault diagnosis method for imbalanced datasets based on MCNN‐Transformer model in industrial processes

Summary Fault diagnosis methods based on deep learning have been extensively applied to the fault classification of rolling bearings, yielding favorable results. However, many of these methods still have substantial room for improvement in practical industrial scenarios. This article addresses the issue of imbalanced fault data categories commonly encountered in real‐world contexts and discusses the characteristics of long time series data in fault signals. To tackle these challenges, a model based on multi‐scale convolutional neural networks and transformer (MCNNT) is proposed. First, in the data processing stage, a diffusion model is introduced to handle the problem of data imbalance. This model learns the distribution of minority samples and generates new samples. Second, the proposed model incorporates an attention mechanism, enabling it to capture the global information of the data during the feature learning stage and effectively utilize the relationships between preceding and subsequent elements in long sequential data. This allows the model to accurately focus on key features. Experimental results demonstrate the exceptional performance of the proposed method, which is capable of generating high‐quality samples and providing a solution to address challenges in practical industrial scenarios. Consequently, the proposed method exhibits significant potential for further development.