F2008-12-168
Vehicle Start-Up Simulation on Drive Train Test Bench and Subjective Comfort Evaluation of Virtual Drive Train by Means of New Driver Modelling Tools Based on Artificial Neural Networks
To be able to meet customer demands concerning comfortability, economical and ecological aspects, automotive industry has turned more and more interests to the subjective comfort evaluation. Many research projects related to driver´s comfort objectification have been carried out for many years by the institute of product development, University of Karlsruhe (IPEK). The goal is to develop a general methodology and different tools to identify dynamic properties of the drive train during start-up, shifting, steering as well as other procedures in early stage of the product development process. The verification is executed by the prediction of the subjective comfort rating. The results of the investigation (i.e. vehicle start-up) have earlier demonstrated that the objective values with the subjective sensation captured during each test can be correlated efficiently. The correlation is achieved by an application of the developed method and the objectification tools based on Artificial Neural Networks (ANNs).
The main purpose of this study is to generate the virtual drive train by transferring the measured data from the drive tests to the dynamic drive train test bench and the simulation models. The advantage for the generation is its ability to partially replace some cost-intensive prototypes and the subsequent drive tests. It is also possible to modify any comfort-relevant parameters in the virtual drive train and follow the upcoming effects by evaluating the comfort rating from the point of view of customers. This article introduces the method of the drive train modelling. An example can be illustrated during a start-up of a front-drive, middle class car. Different start-up characteristic features can vary according to the measured data obtained from the drive tests. Longitudinal acceleration, engine throttle, power spectral density (PSD) and other predefined comfort-relevant characteristic input data can be generated. In this case, the developed clutch system with dual-mass flywheel will be installed on the test bench. Using it in combination with the modified multi-body simulation models, the vibration phenomena like judder and jerking as well as their effects on the degree of comfortability can be investigated.
In addition, the new driver modelling tools will be presented. These are developed on the basis of ANNs from the way individual customers make their assessment. The user-friendly interface of these tools allows both advance users and engineers who are still short on experience in the ANNs field to create different network structures during the training stage. The most suitable network structure for each application can be found automatically. The searching criterion is dependent on the performance of estimation which ranges from 0 to 1. To improve the generalization and the accuracy of subjective comfort prediction, a method and tools to optimise the networks structure based on the sensitivity analysis are further developed. Compared to other conventional methods, this one requires less time. It is possible to calculate the significant value of each input neuron. The redundant ones can then be removed from the networks in order to achieve less computational effort. In the next step, another user interface is used to objectify the comfort sensation from objective data derived from elaborated virtual drive train. The sensation can vary with different customer types according to way of driving (sporty, average and comfortable). Moreover, the developed tools are applied to verify and improve the quality of the virtual drive train. In the long run, by means of the optimization tools, the satisfying comfort ratings should be obtained from the first prototypes.
This abstract is supplemented by a PDF, which can be viewed here.
Poster presentation: Simulation and testing