Research on Multi-Model Fusion Fault Diagnosis Based on Deep Learning

Aiming at the problems of low diagnostic accuracy and slow response speed in diverse fault diagnosis of complex industrial environments, this paper proposes a deep learning-based multi-model fusion fault diagnosis scheme. A novel deep learning model, LADCNN, is proposed, which integrates Convolutional Neural Network (CNN)[1] [1] below, Long Short-Term Memory network (LSTM), and Self-Attention mechanism (Attention). Specifically, LADCNN uses the CNN module to automatically extract local features of input signals, captures dynamic information in time-series data through LSTM, and dynamically weights key features with the Self-Attention module to achieve precise identification of fault patterns. Experiments on the Case Western Reserve University bearing dataset show that the model achieves extremely high accuracy, verifying the effective contribution of each module in improving detection accuracy. This study provides a high-precision and structurally innovative solution for industrial fault diagnosis and offers new ideas for the application and optimization of related technologies.