Combatting the modification of automotive control systems is a current and future challenge for OEMs and suppliers. ‘Chip-tuning’ is a manifestation of manipulation of a vehicle's original setup and calibration. With the increase in automotive functions implemented in software and corresponding business models, chip tuning will become a major concern. Recognizing and reporting of tuned control units in a vehicle is required for technical as well as legal reasons.
This work approaches the problem by capturing the behavior of relevant control units within a machine learning system called a recognition module. The recognition module continuously monitors vehicle's sensor data. It comprises a set of classifiers that have been trained on the intended behavior of a control unit before the vehicle is delivered. When the vehicle is on the road, the recognition module uses the classifier together with current data to ascertain that the behavior of the vehicle is as intended.
A proof-of-concept implementation uses the TORCS racing simulator to generate traces of the engine's behavior. The recognition module extracts features from these traces and feeds them to an artificial neural network (ANN). After training on different tracks, the ANN successfully distinguishes traces originating from the original vehicles as well as traces taken from modified vehicles.
The results show that assessing a vehicle's behavior is feasible and contributes to protect its integrity against modifications. Additionally, the availability of a vehicle's behavioral model can trigger even more interesting applications.