Remaining Useful Life Prediction in Smart Manufacturing Systems Using a CNN-BiLSTM Model with Attention Mechanism

Abstract

With the continuous development of intelligent manufacturing, the maintenance strategy of equipment is also constantly improving, and it is changing from passive maintenance to preventive maintenance and predictive maintenance. Passive maintenance is to perform repairs after equipment fails or shuts down, and this method requires a long downtime maintenance time, resulting in increased maintenance costs. Therefore, this paper combines CNN and BiLSTM to propose an equipment life prediction model, so as to carry out predictive maintenance of equipment through intelligent automation model and improve the prediction accuracy and generalization of intelligent factory equipment RUL. By combining the efficient feature extraction capability of CNN with the sequence data processing advantages of BiLSTM and the weighted redistribution of attention mechanism, the model exhibits excellent performance on multiple data sets. According to the experimental results, it can be seen the advantages of the AM-CNN BiLSTM model are mainly reflected in its high accuracy and stability. On the CWRU dataset, the RMSE value of this model is as low as 0.052, which is better than traditional models, and the prediction accuracy is improved by about 47%. On the UCI dataset, its SCORE value reaches 0.963, indicating stronger generalization ability. All in all, by combining the spatial feature extraction of CNN with the temporal modeling of BiLSTM, and introducing attention mechanism, this model maintains stable performance (fluctuation amplitude<5%) in multi condition data, making it particularly suitable for the analysis and prediction of complex temporal data.