| Mechanical brake is an important safety device for all kinds of transportation vehicles and mechanical equipment,and disc brake is more widely used in practice.The friction performance of brake directly affects its braking performance,which has an important impact on the braking reliability and operation safety of mechanical system.The abnormal change of brake friction state often leads to brake accidents directly.For a long time,research and development of various high performance friction materials and their friction and wear properties and mechanisms have been the main technical direction to improve brake efficiency and reliability.But in fact,any kind of high-performance friction material can not withstand the influence of various adverse external factors indefinitely.Therefore,on-line monitoring of brake friction state and diagnosis and prediction of various faults caused by deterioration of brake friction performance should be an effective technical means to fundamentally avoid the occurrence of brake accidents.In this paper,the phenomena that the braking performance of mechanical brake is significantly reduced or the braking failure is caused by abnormal change of friction state are collectively called "friction failure".The disc brake is taken as the research object,and avoiding the brake accident caused by friction fault is taken as the research goal.The basic theory and technology research on objective characterization of brake friction performance,signal processing of friction state,pattern recognition of friction faults,fusion diagnosis and intelligent prediction are deeply carried out.The research results have important theoretical significance and practical value for improving the working reliability of mechanical brakes and avoiding the occurrence of major malignant braking accidents.Firstly,based on the typical curve of the brake tribological test,the variation rules of the characteristic parameters of friction state,such as friction coefficient,friction temperature rise,friction vibration and friction noise were analyzed.Based on the four characteristics of time,size,change trend and stability,a new type of friction performance characterization parameter which could fully reflect the change characteristics of brake friction process was extracted.A set of 25 parameters of brake friction performance was constructed,which includes three subsets: friction coefficient,friction temperature rise and friction acoustic vibration.Secondly,based on the wavelet analysis method,the signal processing technology which can de-noise,segment and extract features from the friction coefficient signal was studied and established.Based on the characteristics of infrared temperature measurement,the multi-point infrared monitoring method of friction temperature rise and the filtering method of infrared temperature measurement signal were studied.A fast fusion analysis and monitoring model of similar multi-point information of brake friction temperature rise signal was established by using support vector regression machine.Based on the objective characterization parameter set of frictional acoustic vibration and harmonic wavelet packet transform,the processing method of frictional acoustic vibration signal was studied.The internal relationship between friction vibration and noise signals and the friction and wear characteristics in braking process were analyzed.Thirdly,based on the objective characterization parameter set of brake friction performance,the friction coefficient,friction temperature rise,friction vibration and noise signals were refined and synthesized by the method of subjective and objective integrated weighting,and the grade of friction fault was defined and classified.According to the characteristics of brake friction fault and fault data,a multi-class classifier of friction fault modes was constructed,and a model of pattern recognition and fusion diagnosis of brake friction fault was established based on support vector classifier.Aiming at the limitation of supervised learning,the method of friction fault diagnosis based on sparse coding K-SVD algorithm was studied.By learning the dictionary of the original data,the unknown frequency domain friction signal was classified and recognized,and the sparse coding fault diagnosis human-computer interface was designed by using MATLAB.Finally,based on the results of previous studies on friction catastrophe behavior and mechanism of brake,the phenomenon that the temperature of brake friction surface reaches the thermal decomposition temperature of friction material during braking process was defined as a malignant friction fault,and a research idea based on predicting the maximum friction temperature rise of brake to predict its malignant friction fault was proposed.An intelligent prediction model of friction temperature rise of disc brake based on BP neural network was constructed by using the theory of neural network algorithm.Based on the on-line re-learning mechanism of neural network,an on-line prediction scheme for brake malignant friction faults was established.Based on the disc brake simulation brake test bench,a disc brake friction fault diagnosis test system was rebuilt and built.The aforementioned signal processing method and artificial intelligence model were integrated with software,and the on-line diagnosis and prediction tests of the disc brake friction fault were carried out.The theoretical analysis and experimental results showed that the objective characterization parameter set constructed in this paper could comprehensively and objectively characterize the dynamic friction performance of mechanical brakes,taking into account the four characteristics of time,size,changing trend and stability;Wavelet analysis method can effectively reduce noise and extract energy characteristic coefficient of friction coefficient;The multi-point monitoring method and filtering technology based on infrared temperature measurement can effectively monitor and filter the temperature at different locations of the brake disc.Similar multi-point temperature rise fusion analysis model was established by using support vector regression machine.The maximum relative errors of off-line simulation and on-line test were 1.2% and 2.1% respectively,which could better realize the regression prediction of friction temperature rise of brake;The harmonic wavelet packet transform was used to decompose and reconstruct the frictional vibration signals,which could effectively extract the variation characteristics of the frictional signals during braking process.The results were consistent with the friction and wear mechanism of the friction coefficient at different stages;The three main characteristic parameters of friction faults obtained by the subjective and objective integrated weighting method based on expert evaluation and grey correlation degree,and the grade of friction faults defined by threshold method,should be used as effective indicators for pattern recognition of brake friction faults;The classification model of friction fault pattern recognition and fusion diagnosis based on support vector classifier(SVM)achieved more than 95% of the fault diagnosis accuracy of off-line simulation and on-line test,and had high prediction accuracy;The friction fault diagnosis method based on sparse coding K-SVD algorithm not only had a high recognition accuracy for the test data of known fault types,but also could realize the classification and recognition of unknown types of frequency domain friction signals;The intelligent prediction model of friction temperature rise based on BP neural network could predict the maximum friction temperature rise of the current brake more accurately in time scale because the relative deviation between the predicted maximum temperature and the actual maximum temperature was within 6%;Based on online re-learning mechanism,neural network samples and learning experience were constantly enriched.When the sample data were accumulated from 300 groups to 6000 groups online,the accuracy of temperature prediction was improved by about 17%. |
PDF Full Text Request