Research On Remaining Useful Life Prediction Method Of Mechanical Equipment

In the industry production process,mechanical equipment and its parts inevitably wear out and cause failures.How to avoid production shutdowns and safety problems caused by mechanical equipment failures,improve production efficiency,and ensure the safety of workers is a significant problem in the field of mechanical equipment maintenance.The traditional methods are only based on long-time experience,periodically shut down the equipment for maintenance,but such methods would entail additional labor and financial costs.Remaining Useful Life prediction technology of mechanical equipment can effectively solve this problem.RUL prediction technology has important value in equipment maintenance tasks.On the premise that we can predict the RUL of equipment accurately,we can shut down the machine in advance to repair it,this method can not only avoid failures,but also save operation and maintenance costs.Today's society has developed to the era of Industry 4.0,intelligent machines are used more and more widely,and the requirements for their reliability are also higher.This development makes equipment maintenance tasks more difficult and heavy.Obviously,the existing RUL prediction technology based on expert knowledge or physical property will gradually be insufficient to deal with these problems,so it is necessary to research new methods of RUL prediction technology.Data-driven methods have the advantages that it hardly need prior knowledge and it has high efficiency of data utilization,moreover,the data-driven model has better portability.And.The development of Internet of Things technology will provide sufficient data resources for the research.So,research on data-driven methods of RUL prediction is feasible.Therefore,this paper will be based on the data-driven methods,begin research work on new RUL prediction methods.First,this paper introduces the theoretical basis of RUL prediction technology based on the data-driven methons,including the basic principles of typical neural network:CNN(Convolutional Neural Network),RNN(Recurrent Neural Network),etc.,and data preprocessing methods used commonly?evaluation methods of model performance,and other theoretical knowledge.Then,it is noted that the multi-dimensional data collected from mechanical equipment is time series data,and the health status of the equipment has a strong time-dependent relationships,a new deep learning model based on the LSTM(Long-Short Term Memory)and FNN(Feedforward Neural Network)algorithm is proposed.Turbine engine is taken as the research object to carry out research of RUL prediction model.The proposed method it is verified on the CMAPSS data set,and experimental results show that this model has higher prediction accuracy than other algorithms.In order to further increase the accuracy of the predicted value of remaining useful life,this paper continues to optimize and improve the model on the basis of data-driven idea.First of all,using the advantages of one-dimensional CNN in local feature extraction,deep feature mining is performed on the data,and by increasing the dilated rate of the convolution kernel,the receptive field of the convolution kernel is increased while keeping the number of parameters unchanged,thus can cover status information for a longer period of time.In addition,in order to solve the problem of the disappearance of the gradient,a residual block is introduced between the networks,which brings a propagation channel directly to the previous layer for the gradient.Then input the extracted feature vector to the LSTM network to predict the RUL value.Finally,the improved one-dimensional CNN and LSTM network are combined to create a model for RUL prediction.The proposed method was verified experimentally on the data set,and the results showed that the prediction ability of the model was improved.Finally,the shortcomings in this article are summarized,and the future research direction is pointed out.