Remaining Useful Life Prediction Of Industrial Equipment Based On Deep Learning

The rapid development of the industrial level has put forward higher requirements for the safety and reliability of industrial equipment.The prediction of the remaining useful life of industrial equipment is one of the key research contents of equipment prognostics health management.The accurate prediction of the remaining useful life of the equipment can provide a strong basis for the maintenance of the equipment and effectively prevent the economic loss and social harm caused by the sudden failure of the equipment.In the existing research,the research on the remaining useful life prediction of multi-operation equipment and the prediction of lithium-ion batteries with capacity recovery effects are not complete.This paper makes in-depth research on these two issues.Based on the historical sensor data and operating conditions of multi-operation industrial equipment,this paper proposes a remaining useful life prediction model of multi-operation industrial equipment based on dilated convolution.This paper proposes a multi-normalization method for processing sensor characteristics.It converts sensor data in various distributions to the same distribution.At the same time,this paper introduces embedding vectors for feature extraction of operating conditions.This paper proposes a method for processing industrial time-series sequences by using a dilated convolution model,which realizes a direct mapping between sensor data and the remaining useful life of the device.The experimental results on the turbine engine data set and the milling data set show that the model proposed in this paper has better prediction accuracy than existing models,and has good scalability and adaptability.Regarding the prediction of the remaining useful life of the lithium-ion battery with capacity recovery,this paper proposes a fusion method for the prediction model of the remaining useful life of the lithium-ion battery based on the Wasserstein generative adversarial network(GAN).The algorithm uses an artificial neural network model to establish a mapping between the historical battery capacity,rest time between charge and discharge,and the predicted battery capacity.This paper uses empirical degradation models to ensure the correct decline tendency of the battery capacity.Using the discriminator in the Wasserstein GAN model and softmax layer to calculate the weight of the model fusion.The fusion method proposed in this paper realizes the final prediction of the remaining life of the lithium-ion battery by absorbing the advantages of various models.Finally,the experimental results on the lithium-ion battery data set show that the prediction results of the proposed fusion method are more accurate and can better fit the actual battery capacity degradation curve.