Research On Remaining Useful Life Prediction Of System Based On Deep Learning

With the rapid development of the manufacturing industry,the complexity and precision of industrial equipment is increasing,and people have higher requirements for equipment reliability.To improve the safety and reliability of industrial equipment,and reduce the economic loss and personnel safety caused by the sudden failure of the equipment,the industry and academia have done a lot of research in the field of equipment reliability.The remaining useful life prediction is one of the key points.In the actual production environment,the equipment's performance and health conditions are degraded due to unpredictable impact,daily wear,and various corrosion.Remaining useful life prediction uses various methods to analyze equipment degradation data,establish equipment degradation models,and quantitatively analyze the degradation process of equipment,so as to predict the remaining working hours of operating industrial equipment.According to the forecast results,determine the best time for equipment inspection and maintenance,formulate equipment spare parts management and replacement strategies,and maximize the safety of personnel with economic benefits.In this paper,deep learning technology is combined with the remaining useful life prediction,through the data that can reflect the equipment degradation process,the powerful nonlinear fitting function of deep learning is used to solve the task of remaining useful life prediction under time series degradation scenarios.The main work of this paper includes the following points.First,for the remaining useful life in the case that the prediction effect of a single deep learning network needs to be improved,the multi-scale convolutional neural network and the bidirectional long and short-term memory network are effectively integrated,the designed fusion network has stronger feature extraction capabilities.It is verified through experiments that the prediction accuracy of the remaining service life is higher,and the effect is better.Secondly,the remaining useful life prediction problem in time series degradation scenarios is studied.Aiming at the problems of scarcity of degraded data,long collection period and high cost,a method for predicting remaining useful life based on cyclic consistency generative adversarial network is proposed.After experimental demonstration,this method has better prediction accuracy for the remaining useful life prediction task in the scene of scarce degradation data.Finally,it summarizes the initial deficiencies of the above research,and prospects for future research in this direction.