Research On Life Prediction Methods Of Automotive Electromagnetic Relay Based On DBN

In the field of automobile manufacturing,the number of automotive electromagnetic relays used has taken the second place,which play an important role in the control,regulation and protection of the vehicle low-voltage electrical system.However,the working environment of automobile electromagnetic relays is relatively poor,which may be eroded by sand,dust,oil and other pollutants in the car.Therefore,it is of great importance to study the reliability evaluation methods of automotive relays.With the rapid development of electronie teehnology,there are a wide variety of performance parameters at current stage,including contact resistance,loss of quality,spectrum analysis,surface roughness,and so on.There are many existing forecasting methods,such as fuzzy recognition method,grey model method,artificial neural network method and statistical clustering analysis method.Predictive variables selected by these methods are difficult to obtain high measurement accuracy,only the univariate analysis was selected to establish the prediction model which ignored the other factors that affect the relay life,without taking into account the errors and noise introduced into the performance degradation parameters by the external environment and the random of arc erosion and material transfer.To solve above problens,the mainly research work in this paper are as follows:1.In order to achieve the goal of predicting the life of automotive electromagnetic relays by using a hybrid algorithm,the reliability evaluation and life test system of automotive electromagnetic relays was designed and implemented to complete the acquisition and preservation of performance degradation parameter data.The sources of noise in performance degradation data of automotive relays were analyzed,moreover the kalman filter algorithm was used to denoise and smooth the datas obtained by the experiment.2.Considering that the performance parameters timing values of the relay are non-stationary time series,this paper proposed a prediction method with multiple performance degradation parameters based on the Deep Belief Networks(DBN).Six key performance degradation parameters such as super-path time,boxrnce time,contact resistance,pick-up time,release time and arc time were used as the inputs of the neural network,the relevant features were extracted by the hidden layer nodes of the neural network,residual life value was the output of the neural network.The experimental results show that the six performance degradation parameters were easy to be measured and the prediction accuracy is relatively high.3.In order to overcome the shortcomings of artificial neural network,such as slow convergence rate,easy to fall into local extremum and easy to overlearming,this paper proposed an improved cuckoo search algorithm that introduces the patrol factor and local chaos search principle,and combined the algorithm with the deep belief network to form a new hybrid prediction algorithm to obtain better prediction results.The hybrid prediction algoritiun with the combination of improved cuckoo search algorithm and deep belief network,the hybrid prediction algorithm with the combination of cuckoo search algorithm and deep belief network,and the deep belief network prediction algorithm were compared,respectively.The experimental results showed that all of the three prediction algorithms can realize the life prediction of automotive relays,and the method with the combination of improved cuckoo search algorithm and DBN had the highest prediction accuracy and the fastest convergence rate.1-6 key performance degradation parameters were selected from the six key performance degradation parameters for comparative study,and the influence of different key performance degradation parameters on the prediction effect was verified.By comparing with Wavelet Neural Networks,Radial Basis Function Neural Network and Elman Neural Network,the superiority of the prediction model proposed in this paper was verified.