Analysis of an e-scooter and rider system dynamic response to curb traversing through physics-informed machine learning methods

Abstract

The use of big data to statistically evaluate system performance with the help of Machine Learning (ML) tools is a very popular method. Classic experimental methods of system performance can be impractical when needing to test a variety of input parameters within a range of specific limits and can take enormous time and effort. Mathematical models which numerically mimic the physical system dynamics can be deployed for the same aim. In this study, a system modeling based data analysis approach has been implemented to gather the data to understand the input-output relationship of an e-scooter and rider system in terms of its stability dynamics while driving over a curb of a specific height. Based on the test conditions specified in the German e-KFV standard, a virtual driving test has been created which consists of a kinematic e-scooter model with its rider and a test track. In the range of input parameter space, over 20,000 numbers of different cases have been solved and the results are recorded. The interaction between the rider and the vehicle is achieved through elastic connections which enable not only the recording of loading data on the arms and legs, but also allows the simulation to reflect the real-time effects of the rider scooter interaction. Through statistical analysis and using ML techniques, the influence of each parameter (tire size, speed, rider mass, etc.) on the objective outputs (loading on arms, ability to ascend the curb, etc.) regarding the stability and safety while driving over the curb have been reported and shared for further analysis by the e-scooter community.