Study On Method For AMT Shift Quality Evaluation Based On Neural Network

Automatic Mechanical Transmission is non-power shift device, which produces jerk during gearshifts because of the torque interruption. So, the smoothness performance of AMT is worse than that of other transmissions. The research on shift quality evaluation is the key to develop AMT products. How to build an evaluation system on shift quality is an important research topic in order to inprving shift quality. And an excellent control theory for gearshifts is based on a comprehensive evaluation system.Today there exist no specific objective grading methods for shift quality. Existing methods usually give a common grading for the entire vehicle driveability, and not the grades for specific subsystems. One reason is that the factors are many and it is not obvious how these affect the total shift quality impression. And another reason is mainly that objective evaluation tools are developed by commercial companies that, of course, do not reveal the details of their methods. When people in vehicle industry perform shift quality evaluations, a scale from 1 to 10 is normally used, based on the subjective assessments of experienced test drivers recorded during a sequence of relevant manoeuvres is usually employed. Each driver has his or her own perception of vibrations and thus their own subjective evaluation scale. Though subjective ratings of qualified evaluators are important for vehicle development, they are generally inconsistent, and make comparison of different shift events difficult; to say the least. Little definitive work has been published which correlates shift-specific characteristics with objective ratings.The goal of this master thesis,as is a part of Study of Dynamic Comprehensive Evaluation System on Shift Quality of Vehicle project, is to develop a grading system for Shift Quality Evaluation based on network which is compared with professional test driver's subjective grading.Vehicle shift quality is determined by many complex factors, and it will be subjected by driver's intent and sensation. The course of evaluation is dynamic and nonlinear. BP network can approximate every nonlinear function, so it can be used to evaluate the shift quality of vehicle.In this thesis, the method has been evaluated on real data measurements collected using a test AMT car. The objective data were acquired during test drives using data acquisition equipment and the subjective data were gained by having the test drivers fill in questionnaires about their perception of the car. The method is implemented in Matlab/Simulink and a specially designed objective evaluation modle has been constructed, which enables the possibility of producing ratings. A graphical user interface (GUI) in Matlab is used to control the program and its output. The system has been named to'Vehicle Objective Evaluator of Shift Quality'. So work was undertaken towards a software tool to predict subjective ratings of shift quality from powertrain and vehicle data. The objective evaluation model of shift quality is a muti-input and single-output system. The trained neural network instead of human beings is used to evaluate shift quality. The shift quality evaluation indices are inputted to the input layer of network, and through the identification of the network, the evaluation ratings output automatically. The math model of the relationship between shift quality evaluation rank and evaluation indices can be established as: R = f ( a , j , t ,Δωe), where a is shift acceleration; j is jerk; t is shift time andΔωe is fluctuation of engine speed. [ a , j , t ,Δωe] is a 4-dimensional input vector. According to different vehicles, the parameters might be adjusted. It is important to point out that the selection of shift quality evaluation indices must be reasonable and well-founded. These indices must be proved to be significant factors affecting shift quality , otherwise, they might produce negative influences to evaluation ratings.Using artificial neural network to evaluate AMT shift quality has good digital approximation and stability. By comparing and analyzing evaluation results, it is shown that this objective method can evaluate shift quality effectively and the results havegood consistency with subjective assessment. Therefore, the objective evaluation method based on artificial neural network is proved to be reasonable and promising, which is also a theory base for developing advanced neural network with on-line learning function.