Ease Of Control Evaluation Via Stochastic Driver Model

Nowadays, with the vast development of automobile industry, more and more citizens areable to afford family cars, leading to diversification of the driver population. Therefore,vehicle designers should not only satisfy the requirement of professional drivers tomanipulate, but also give adequate consideration to those who are not well-trained,especially the elder ones. It is the engineers`duty to design such vehicle that can be handledby the majority without effort. Ease of control in this paper is the ability of vehicle to adaptto the crowd. In another word, it is the measurement of the size of the population who can, indaily driving behavior, easily control the designed vehicle, under the condition of certainclosed-loop performance to some extent.Vehicle design should take these points into consideration: relieving the driving burden,enhancing the ride comfort and improving the safety performance. The challenge ofanalyzing and designing vehicle handling performance is always the major research directionof worldwide professors and engineers. The obvious feature for analyzing the vehiclehandling performance is the necessary involvement of drivers; cause vehicle and driverinteract deeply. Not only the quickness and precision of vehicle response should beconsidered during handling performance analysis, but also the response properties of driversto the vehicle state. In short, vehicle and driver as in the close loop could not be separated,because they interact deeply under the domain of human.Currently, there are mainly three methods of analyzing vehicle handling performance,⑴subjective evaluation which carried out by skilled drivers during verification and tune phase, ⑵objective evaluation in proving ground with test instrument vehicles,⑶objectiveevaluation with simulation methods during definition design and verification. However,subjective evaluation is the final sign. The vital disadvantage of subjective evaluation is lackof coherence. Objective evaluation has its advantages: explicit experiments standard andindicators and without human being uncertainties. Furthermore, with profound developmentof virtual simulation technology and accurate construction of vehicle model, analysis andpredicting the vehicle dynamics based on computer simulation are much easier to realize,and prediction of vehicle handling performance could be made before prototype vehicle orduring benchmarking stage, leading to optimization design with cost and time saving.The research objects of vehicle ease of control include the driver crowd and vehicles. Asvehicle handling characteristics have the features of diversification and randomization, for aspecific vehicle, the input of control is of randomness and uncertainty within a certain range.For this character, in the actual analysis of vehicle ease of control there are two mainmethods. One method is by well-trained drivers through lots of experiments in simulator orin prototype vehicle, which requires expensive test hardware, such as prototype vehicle anddriving simulator, and consumes insupportable labor and time costs. The other one is bydigital driver model through virtual simulation analysis and prediction in computer, whichcan be very efficient and cost saving as the simulation experiment can be realized bycomputer software, according to a specific vehicle model from the data base of handlingperformance of the driver crowd. The data base is constructed through experimental researchin advance.Therefore, the aim of this thesis is mainly to explore a closed-loop approach to analysisvehicle ease of control, expecting to carry on the analysis completely through digitalsimulation, instead of lots of real experiments by fully-trained drivers. And this thesispresents a quick, uncomplicated and cost saving method of constructing simple vehiclemodel through system identification technology in benchmarking phase, which could furtherexpand the use of computer simulation technology in automobile research and development. Based on extensive reviews, the following questions were studied:First, because there exist nonlinear element, complex co-couple between the outputs existswithin multiple-input multiple-output vehicle system, it is not easy to use systemidentification technique to set up vehicle model. It is necessary to make some suitableassumption for the nonlinear problem. Furthermore, considering vehicle speed as input couldnot be a persistent excitation and the relative simple vehicle model could meet the need inbenchmarking phase, a simple but fast vehicle modeling procedure via system identificationtechnique was proposed and the transfer function of the2DOFs vehicle model was identified.First, the single input single output of the vehicle model was identified as the steering angleis the input. According to the transfer function property, vehicle speed was separated asindependent variable in the yaw rate to front steering angle and Latac to front steering angletransfer function. To achieve the data, steering wheel angle pulse test at the speed of100Km/h was carried out, and the vehicle model via the given identification proposal was setup and comparison with the test data from steering wheel angle step test at different speedand lateral acceleration. The coherence between them leads to the validity of the method.This method is simple and easy to carry out and meet the pre-development stage in casethere is no detailed specification of vehicle parameters, and since it could derive the majorvehicle dynamics properties for analyzing, it is an efficient and cost-effective method to setup vehicle model.Fourier transform method and the ARMA model were used to identify the single input singleoutput vehicle system respectively. When noise exists in the proving ground test data,Fourier method result in low accuracy, because of the inevitable noise in the field test,steering wheel angle is not zero, the presence of residual zero drift, etc. Truncation andleakage problems inherent with Fourier transform when processing data make frequencyresponse characteristics of the estimation accuracy is low. Whereas the ARMA modelmethod could derive a smoother curve of frequency response and without the problem ofinsufficient frequency resolution. Due to its simple and consistency, ARMA model method is a promising tool to derive the frequency response in engineering practice. In addition,since it is a kind of parameter identification method, the results could be used to get theidentified2DOFs vehicle model with speed as an independent variable.Second, in order to accurately describe the driver's behavior, identification of driver modelparameters through experiment should be carried on. Only when the car was controlledunder drivers, the identification could be done. Because drivers' perceptual function as thefeedback loop could not be cut off during driving, it is closed-loop system identificationproblem. In accordance with the closed-loop system identification theory, to use indirectclosed loop identification approach, the driver model should be set up first, and thentransform the closed loop to open one to identify. Based on extensive study on the drivermodel theory, considering the Position and Orientation optimal preview driver model (POdriver model) has the advantage of simple and clear physical meaning, and accuracy forfollowing great curvature path, the PO model was chosen as the identified driver model. Theidentified parameters were classified into2categories: structure identification and parameteridentification. The structure parameters Td and Tp were used for identify the otherparameters during parameter identification process; whereas the parameters derived fromparameter identification process provide optimization direction for the structureidentification algorithm. The difference between the sample points of frequency responsefrom ARMA method and that from the identified parameters driver model was minimized byLevenberg-Marquardt algorithm to fit the physical parameters of the PO driver model.Third, ease of control evaluation via stochastic driver model through simulation wasexplored. In this paper, extensive research over the worldwide range for closed-loop systemperformance analysis was reviewed, and was divided into three categories:⑴based on theevaluation of the optimal closed-loop performance;⑵based on the perceptualcharacteristics of the driver physiological or modeling parameters;⑶analysis of driver'sburden as indication of closed-loop performance. Obvious feature of this thesis isconsidering the driver diversity and randomness; the driver's control inputs with randomness and uncertainty cause the closed-loop system randomness as well. Therefore, first of allstochastic driver model was built based on statistics, and then closed-loop performanceindicators and safety metrics were set up. Using Monte-Carlo method, statistical result of theclosed-loop response under random driver input was analyzed. By comparing theundersteeing car, neutral steering car and oversteer car, the random closed loop performancewere compared, and indicate that the method does work well for the ease of controlevaluation.Above all, this thesis has the following innovation points:1. Proposed quick and easy vehicle model identification method through studying thecharacteristics of the2transfer functions derived from2DOFs vehicle model.2. Using Monte Carlo method, ease of control evaluation was explored by sampling therandom driver parameters.