The Study On Embedded Condition Detection And Fault Diagnosis Technology For Chassis Of Self-propelled Gun

Self-propelled gun's chassis system consists of power installation,actuating device and moving device.Its performance is decisive for gun's power and maneuver capability.And the technical characteristics of high density, multi-function and integration make its maintenance very difficult.The traditional detection method for self-propelled gun's chassis system is on-line detection without disassembling.The defection of this method is that equipment is only detected at zero loads.Because there is no dynamic load,useful information reflected faults in the tested signals are often concealed by noises.And even sometimes there is no useful information in the tested signals.So,the fault location accuracy of the traditional detection method is in a low lever.And as the small sample characteristic,it's used to lead some fault diagnosis methods based on traditional statistics theories,such as neural network,to over-learning,low generalization and local minimum when training and recognizing categories of faults.In this paper,embedded detection system based on wireless distributed networks was designed to avoid wire communication and make it possible to detect chassis system under load.Sensor groups got information from important parts of chassis system,and then embedded detection system collected the information and transformed it into digital data,which was ultimately transmitted to the centre of signal analyzing through wireless distributed networks.Two kinds of methods were used to extract the characteristics of engine vibration signals.One was dimensionless index of magnitude domain,extracted by microcomputer in embedded detection system.The other was feature vectors based on energy of frequency bands.By means of the distribution characteristic of signal energy in the wavelet packet space and the method of self adapting denoise,eigenvectors were extracted.To solve the small sample problem in fault diagnosis,support vector machines (SVM),a method based on small sample theory,had been applied in this paper.And fuzzy theories were used to improve the performance of SVM multi-fault classifiers.In order to overcome the problem that SVM multi-fault classifiers could not recognize unknown fault patterns,the modified ART2 neural network was utilized.