Research On Prediction Technology Of Equipment Remaining Life Based On Deep Learning

As the complexity of mechanical equipment increases,the stability and reliability of its key components are of great significance to the operation of the equipment.Therefore,it is very necessary for the health management of key components of complex equipment.However,because key components are usually complex in structure and operating conditions,how to ensure their safe and stable operation has become the core issue of equipment operation and maintenance.Traditional methods such as fault repair and regular maintenance can easily cause time cost,manpower and material resources consumption.,Reduce the efficiency of the equipment..In recent years,the emergence and rapid development of fault prediction and health management(Prognostics and Health Management,PHM)technologies have provided strong technical support for the improvement of maintenance methods and maintenance strategies.Among them,remaining useful life(RUL)is used as PHM The main core of technical research has also become the focus and focus of current research in the PHM field.The complexity of RUL technology has also made it a difficult point in the research of PHM technology.At present,with the rapid development of big data and artificial intelligence technology,data-driven RUL technology finds rules and builds models from a large amount of equipment operating data,does not need to establish precise mechanism models,and has the advantages of low cost and obvious effects.It has become The research hotspots of PHM technology for mechanical equipment are discussed.Among the many data-driven RUL technologies,the methods and strategies based on deep learning aim at the characteristics of mechanical systems that are difficult to establish statistical models and mechanism models.Deep neural networks are used to extract the in-depth characteristics of faults from equipment historical fault information,which reduces the model establishment.It is difficult and has the characteristics of good predictive effect,and is currently widely used in RUL.This paper studies the remaining life prediction of equipment based on deep neural networks,and develops related application systems.The main work is as follows:(1)Combined with the requirements of RUL,the traditional deep neural network structure has the problem of low prediction accuracy caused by insufficient deep feature extraction.The network structure is improved and optimized,and a variety of network structures are used.In this way,multiple network structures such as RNN,CNN,LSTM,etc.are combined and applied,and a new network combination model is proposed,and experiments are carried out on bearing data.The experimental results show that the prediction effect has high accuracy.(2)In order to further improve the prediction accuracy,in order to solve the problems of many parameters and time-consuming in the training process of the RUL prediction model,the independent recurrent neural network(INDRNN)structure was introduced on the basis of the network combination model proposed above,and the network parameters were analyzed.Reasonably set up,and reasonably standardize the number of training,in order to achieve the best prediction effect.Experiments on the aerospace engine data set show that the new integrated network structure based on INDRNN can not only improve the prediction accuracy,but also improve the efficiency of model training.(3)Based on the previous research,this paper conducts RUL model training based on the Tensor Flow framework,and develops an equipment remaining life prediction system on this basis,and realizes the remaining life prediction based on equipment operating data,which can change the maintenance method of the equipment from the traditional Passive maintenance is transformed into active maintenance,which reduces the frequency of equipment maintenance and improves the safety performance of equipment operation.