Prediction Of System Remaining Useful Life Based On Deep Learning

As the key and premise of intelligent maintenance decision of complex equipment system,remaining useful life(RUL)prediction has gradually become a popular direction.With the development of big data and artificial intelligence technology,data-driven technology has achieved good results in the application of equipment health management.Deep learning has the advantages of strong learning ability,wide coverage and strong adaptability,and has become a major prediction method.In this paper,the prediction models of the remaining useful life of the system are established based on deep learning method,and the data set of the turbofan engine published by National Aeronautics and Space Administration is analyzed.The main research work and results are as follows:(1)Based on the complexity of data degradation,a temporal convolution network model based on piecewise nonlinear degradation is proposed.In this model,a piecewise nonlinear target function is used to describe the trend of data degradation.A temporal convolutional network with Mish activation and a full connection layer are used for feature extraction and life prediction,respectively.Experiments were carried out on the C-MAPSS dataset,and the effects of batch size and other parameters were studied to determine the optimal parameters of the model.Experiments show that the Mish activation can improve the nonlinearity of the model,the model has a lower error than the model whose target function is piecewise linear function,and is better than some existing prediction methods.(2)Considering the potential relationship between different index data,a multiscale adaptive attention network model is used to predict the remaining useful life.In this model,multi-scale convolution is used to extract the features of the data,and then adaptive attention module is combined to fuse the feature relations between the data from vertical and horizontal dimensions respectively,so as to predict the remaining useful life of the full connection layer.Empirical analysis was carried out for the turbofan engine data set,and the optimal model was established by comparing the influence of the number of adaptive attention modules and other parameters.Experimental results show that the prediction accuracy of the model can be improved by using multi-scale convolution and adaptive attention module,and the validity of the proposed method in predicting the remaining life of complex systems is verified.The above two models can accurately predict the remaining useful life of complex systems and are better than some existing prediction methods.The experimental results also prove the advantages of nonlinear target function,nonlinear activation,multi-scale feature extraction and adaptive attention module,which provides a new idea for studying the remaining useful life of other complex systems.