Abstract
For the statistical simulation of longitudinal dynamics of vehicles with manual transmissions, a statistical driver model (SDS = Statistical Driver Simulator) has been developed at the IfF of TU Braunschweig. The inspiration for the development of the tool came from the necessity to simulate different driving styles of different drivers under continuously changing traffic environments, which is the prerequisite for the simulation of realistic load spectra for drive train components. The statistical modelling of the driver behaviour in SDS was gained from the analysis of road measurements with 20 test drivers in different vehicles on more than 60,000 km of public roads. The measurements were utilised for the determination of the statistical driver behaviour, which is the basis for drivers actions in SDS. The driver behaviour modelled in SDS is discussed in detail incorporating the models for vehicle speed, accelerations, decelerations, drive offs and gear changes.
During the simulation run, the driver must be given a course of changes in desired vehicle speed. SDS uses a 'step function' called orientation speed profile (OS), which is the vehicle speed the driver is trying to reach by accelerating or retarding the vehicle. By jumping continuously during the simulation run, the OS provokes the natural driving process. Each driving style and road has its characteristic OS distribution which has been extracted from measurements under predefined conditions.
The OS distributions also contain, besides the characteristic road features such as the number of curves and speed limits, the influence of traffic volume at the time of the measurement. Therefore a separate simulation of the flow of traffic is not necessary in SDS.
Due to the considerable influence of gear changes on the input loads of manual transmissions, an extensive model for the simulation of gear changes with manual transmissions had to be developed. The set-up of the gear change model is explained in detail. In order to be able to depict the natural variance of various drivers with different drive styles in the model, the measured gear changes were transformed into parameter sets and stored in a statistical data base (comparable to the parameters in ECUs of automatic transmissions). With the road measurements, numerous parameter sets of this type were generated for all different types of gear changes (1-2, 2-1, 2-3, 3-2, etc.) with predefined driving styles, traffic environments and vehicle weights. In the model each type of driver action is generated by situation-dependent statistics.
This situational dependence is achieved by the utilisation of multidimensional statistical distributions, which means that each parameter is statistically generated in dependence of one or more input parameters. If this situational dependence is neglected in a statistical driver model, this will invariably lead to a chaotic driver behaviour. Examples for multidimensional statistical distributions are given in the paper. As a possible application for the SDS driver model, the influence of the DRVparameter space (consisting of driver, road and vehicle properties) on the service life of drive train components is shown with the roller bearing in a manual transmission as an example. A diagram shows the service life of the roller bearing as a function of driving style, road type, vehicle weight and engine power. With this type of diagram it is possible for the first time to judge the different parameters of the DRV-parameter space according to their importance on the service life of transmission components under customer operation.
Significance to the progress of automotive engineering:
In conjunction with the service life analysis in the DRV parameter space, SDS enables the user to simulate tailor-made customer load spectra for drive train components (especially manual transmissions). The vehicle producer gets the opportunity to identify possible light-weight potentials and weaknesses in the design of the new drive train even before the first prototype has been built. With the knowledge of the representative load spectra, existing test procedures can be optimised or new tailor made test programmes be determined. The risk that cost and time consuming work is iterated in the design process at a late project phase can be reduced effectively.