Research on Fault Diagnosis and Prediction Model of Intelligent Manufacturing System Based on Deep Learning

Traditional fault diagnosis methods struggle to meet the requirements of real-time and accuracy. Deep learning (DL) algorithms, due to their advantages in automatic feature learning and processing high-dimensional nonlinear data, have become a key technology for addressing this issue. This study employs a CNN-LSTM hybrid neural network model that combines the spatial feature extraction capability of Convolutional Neural Networks (CNN) with the time series analysis capability of Long Short-Term Memory networks (LSTM) to conduct deep analysis on multidimensional time series data from Intelligent Manufacturing Systems (IMS). The data is collected in real-time by sensors and stored in the system database. Data preprocessing involves cleaning anomalies, standardization, and normalization to meet the training requirements of DL models. The CNN-LSTM model extracts local features through multiple convolutional layers, then captures long-term dependencies using LSTM layers, and finally performs classification or regression tasks via fully connected layers. Experimental results show that the CNN-LSTM model excels in fault diagnosis accuracy, recall, and F1 score, achieving an accuracy rate of 96.83%. Fault prediction results indicate that the model has high prediction precision and stability, with a Mean Squared Error (MSE) of only 0.02 hours.