Driver Classification of Shifting Strategies Using Machine Learning Algorithms 2020-01-2241
The adequate dimensioning of drive train components such as gearbox, clutch and driveshaft presents a major technical task. The one of manual transmissions represents a special significance due to the customer’s ability of inducing high force, torque and thermic energy into the powertrain through direct mechanical interconnection of gearstick, clutch pedal and gearbox. Out of this, the question about how to capture behavior and strain of the components during real operation, as well as their objective evaluation evolves. Furthermore, the gained insights must be considered for designing and development.
As a basis for the examination, measuring data from imposing driving tests are adduced. Therefore, a trial study has been conducted, using a representative circular course in the metropolitan area of Stuttgart, showing the average German car traffic. The more than 40 chosen drivers constitute the average driver in Germany with respect to age, gender and annual mileage. The used vehicle is equipped with high resolution data acquisition in order to determine inner systemic variables such as oscillation of the drivetrain, clutch slip, engine and wheel torque as well as outer variables such as pedal stroke and acceleration during the shifting process, which the driver controls. Misuse aspects resulting in increased wear or even component damage as well as comfort-related aspects are collected and consulted for further development.
In the context of this study the collected measurement data are analyzed in detail. On basis of this analysis, an evaluation of the shifting processes during real operation is conducted and presented by use of objectified characteristic variables. Finally, the evaluations of different shifting processes are opposed. With the help of t-SNE and cluster analysis as a classification method on basis of a machine learning algorithm, different types of drivers can be worked out and described in detail using the determined characteristic variables.
Citation: Trost, D., Ebel, A., Brosch, E., and Reuss, H., "Driver Classification of Shifting Strategies Using Machine Learning Algorithms," SAE Technical Paper 2020-01-2241, 2020. Download Citation
Daniel Trost, André Ebel, Erwin Brosch, Hans-Christian Reuss