Study On Fault Diagnosis And Tolerance Technology Of AMT System

Electronic technology in vehicle integrates auto technology with electron technology. So, as technology of auto and electron developing, the application of electronic technology to vehicle becomes more and more extensive and the electronic grade of vehicle becomes higher. Automatic mechanical transmission (AMT) is the outcome of the development of electronic technology. It greatly improves the convenience of vehicle and reduces the intensity of drivers'work and satisfies drivers'requirements.Because of the use of electronic technology on AMT, automatic gearshift comes true. At the same time it advances the development of on-board diagnosis and tolerance technology of AMT. In this paper we mostly study on-board diagnosis and tolerance technology of AMT system.Now more and more diagnostic technologies of dynamic system are used to diagnose the faults of vehicle. For example, traditional methods as equipollence space method are used in fault diagnosis of engine. Also the branch methods are used in engine diagnosis. More and more researchers and scholars begin study the use of many advanced diagnostic method for fault diagnosis of vehicle electronic system such as detecting filter method, dual-sect detecting filter method, adaptive non-linear observation method and irregular builder method. Current diagnostic methods of auto electronic system are some methods which use the changing range and changing rate of signal-self. But these methods have larger limitation and lower sensitivity. For some components with high requirement, these methods are not adaptive. So in that case, we need more advanced diagnostic methods. Many specialists and scholars also present many on-board diagnosis methods of the components of control system of automatic mechanical transmission. Such as, the method which use direct redundancy and dynamic redundancy relationship among the components in electronic system of AMT to diagnose the fault in feature or in essence; the method which use the part model of transmission assembly of vehicle and the logic relationship of some components to diagnose the faults of AMT control system on-board; the diagnostic method of electronic system of AMT which is based on model, and so on. In this paper, diagnostic method based on BPnetwork is used to diagnose faults of AMT control system. And the diagnostic method based on logical relationship of components is also used in this paper. For the fault diagnosis of pressure sensors and temperature sensors, diagnostic methods based on extremum are used.According to frame of electronic system, the faults of AMT control system contain: faults of transmission control unit (TCU); faults of actuators; faults of sensors and faults of control objects. And according to frame of AMT system, the faults contain: faults of components related to clutch; faults of components related to gearbox and faults of engine components. Also the control system about tolerance contains: TCU tolerance system; shifting actuator tolerance system; throttle control unit tolerance system; clutch actuator tolerance control system and sensor tolerance system. This paper has designed appropriate tolerance methods for each part, which are based on characteristic of the faults of every part. These methods enable the car with faults to arrive at his/her destination or Repair Bay safely.There are many advancements of fault diagnostic methods used neural network as follow:(1) Neural network not only can be trained off-line but also can study more on-board, in this way it can get more accurate fault model.(2) Neural network can treat with very complex non-linear problems.(3) This method has functions about signal pattern switch and character pick-up.(4) Less sensitive to system containing indeterminacy and noise, and to the instance when input pattern is not perfect.(5) Decision-making does not need statistic model. This point has application value in engineering.There are five kinds of fault diagnostic methods using Neural Network: the first one uses neural network to diagnose fault directly. The second one uses neural network to produce error. The third uses neural network to estimate error. Next is using neural network to diagnose further. And the last one uses neural network to compensate adaptive error.Key designs of diagnostic network contain confirming the number of layers, the number of neurons, the number of epochs, and selecting the initial weights and the training data. Because of the limit of trial condition, the training data arecollected using MATLAB model. The number of layers and neurons are selected with try-collect method. The first step of this method uses only one concealed layers and less neuron. Then increase the number of neurons of concealed layer, until advanced performance has been got. Otherwise use two concealed layers.The structure of fault diagnosis system using neural network in this paper is 6-40-5. This paper validates feasibility and validity of the diagnostic method using neural network by building model and simulating.