Research On Testing Method Of Moisture Content Of Brake Fluid Based On Electrical Parameters

Brake fluid is the medium that transfers the brake pressure in the hydraulic brake system,and plays an important role in the safe driving of the car.The detection of the brake fluid moisture content is an important technology to ensure that the brake fluid is qualified.Therefore,research on a method that can quickly and accurately detect the moisture content of brake fluid has important theoretical research significance and use value.This paper first designed the lower computer and the upper computer of the water content detection system,and verified the reliability and correctness of the detection system through experiments,and completed the collection of electrical parameters of brake fluid through experiments.Using the collected electrical parameters,three machine learning methods,such as BP neural network,support vector machine and extreme learning machine,were used to construct a nonlinear mapping relationship between the brake fluid moisture content and the electrical parameters,and the completion of the brake fluid Prediction experiment of water content.The experimental results show that the overall mean square error of the regression prediction of the BP neural network is 0.002,the determination coefficient R value is 0.9574,and the classification accuracy is 86.1%;the overall mean square error of the regression prediction of the support vector machine is 0.0043,the determination coefficient R value is 0.9662,classification The accuracy is94.4%;the regression mean prediction value of the extreme learning machine is 0.009,the determination coefficient R is 0.8778,and the classification accuracy is 88.9%.Furthermore,a long-short term neural network and a stack sparse autoencoder were used to complete the prediction experiment of the brake fluid moisture content.The experimental results show that the regression prediction of the long-and short-term neural network is roughly a straight line,and the moisture content of all samples is predicted to be a value,and the samples are divided into one category,indicating that the long-and short-term neural network is not suitable for the analysis of such data.After SAE optimization,only the prediction effect of the extreme learning machine becomes better,the mean square error becomes smaller,and the classification accuracy becomes higher;the prediction effect of the BP neural network becomes worse,the overall mean square error becomes larger,and the goodness of fit becomes smaller,but The classification accuracy has been improved;the support vector machine prediction effect has seriously deteriorated,the fitting goodness is 0.0949,the mean square error has also become larger,and it is almost impossible to predict the water content.After expanding the data by spline interpolation,the prediction effects of the three models have been improved.The mean square errors of SAE+BP,SAE+SVM,and SAE+ELM are 0.0081,0.0159,and 0.0004,and the classification accuracy is respectively 96.3%,88.9%,95.4%.From the experimental results,we can know that the SAE+ELM model has the best regression prediction effect,and the SAE+BP model has the best classification effect,and the accuracy rate reaches 96.3%.It shows that the detection system can meet the requirements of brake fluid moisture content detection.