Fault detection of production line based on improved deep belief network

The efficiency of production lines has a significant influence on modern manufacturing production. Fault prediction can remarkably enhance the stability and efficiency of production lines. However, due to the impact of equipment operating status, operating characteristics, and fault features on production line efficiency, fault prediction poses a challenge. In recent years, numerous machine learning and deep learning methods have been employed for intelligent fault detection. The present work proposes an improved deep belief network (DBN) model that integrates an attention mechanism. By utilizing the Monte Carlo method and Bayesian evaluation to fuse the collected fault features of the production line, and combining the attention mechanism with the layer-by-layer greedy learning algorithm to extract deep fault features from the monitored equipment operating status information, backpropagation is used to fine-tune the parameters, thereby enhance the fault recognition ability and constructing an improved DBN model. The results demonstrate that the proposed model has a faster convergence speed and an average accuracy of 96.7% in fault prediction, which can better satisfy the requirements of factories for production line fault detection, ensuring that faults can be quickly and accurately identified and providing support for the production field.