Research On Remaining Useful Life Prediction And Anomaly Detection Of Industrial Equipment Based On Deep Learning

With the rapid development of technologies such as the Internet of Things and artificial intelligence,smart manufacturing has become an important approach for the digital transformation of the manufacturing industry.Prediction and Health Management(PHM)is a critical method to realize intelligent manufacturing.It performs the remaining useful life prediction and anomaly detection of equipment through data analysis,artificial intelligence,and other technologies.Traditional knowledge-driven methods can hardly deal with the emerging challenges in the era of industrial big data,and data-driven deep learning methods are gradually gaining popularity in the industry.This thesis mainly focuses on the two core contents of PHM—the remaining useful life prediction and anomaly detection of equipment.The main contents are as follows:(1)For the prediction of equipment’s remaining useful life,this thesis proposes a novel prediction model AA-LSTM based on adversarial autoencoders.The model uses a Bi-LSTM based adversarial autoencoder to extract degradation information in time series.In addition,the model can jointly optimize multiple modules through the proposed joint training algorithm.We evaluated the performance of AA-LSTM on the turbine engine dataset,which shows clear advantages over current state-of-the-art methods on complex datasets.(2)For the multivariate time series anomaly detection,this thesis proposes a Transformerbased multivariate time series unsupervised anomaly detection algorithm TFCAD.The model extracts the temporal correlation of data and the correlation between features through the time series relation extraction network,and the discrete relation extraction network automatically learns high-order cross information between discrete features.We evaluated TFCAD on two public datasets,and it outperformed current state-of-the-art models with interpretability to some extent.