Detection of distinctions in car fleets based on measured and simulated data

The main aim of this paper is the analysis of a fleet behavior and comparison of damaged and non-damaged vehicles using different machine learning algorithms to identify the main distinctions in a car fleet. For this purpose OBD (On Board Diagnostics) signals are recorded in three various car types (BMW X3, Peugeot 208, Seat Leon) and several hundred drives. Furthermore, thousands of drives are simulated based on the Discrete Fourier Transformation (DTF) and Monte Carlo simulation (MC) to provide a huge training data for the rule learning algorithms. The simulated data represents the car fleet with given (or assumed) testing conditions. For example it is assumed, that all the cars of the fleet don't exceed the maximum speed of 170 km/hour due to the specs of the product. All made assumptions are variable and can be changed after the analysis process. All signals (simulated and recorded) are analyzed in order to provide information about every single drive (e.g. out of the speed signal: max speed, max acceleration, duration of the drive are determined and stored in a data base).