A novel framework for vehicle functions identification by exploiting machine learning techniques

Nowadays vehicles architectures exploit various automotive network protocols that bring information between the implemented Electronic Central Units (ECUs). Exchanged data are encoded and only Original Equipment Manufacturers (OEMs) and T1 (Tier One) producers know their meaning and how decode them. A software model will be developed in order to detect vehicles functions without having database files associated to network signals. Furthermore, the model will behave like an ECU by producing output signals related to input ones. Machine Learning techniques will be exploited, in particular Clustering task will be exploited to understand not a priori known vehicle functions and a Neural Network will be implemented to emulate an ECU behavior. Signals will be grouped in five different types of vehicle functions and the model will predict the ECU’s output data with high accuracy. Applications concerning the developed project are, in primis, to fix up possible vehicles electronics faults. In addiction, vehicle predictive maintenance could be done. Another application, could be to check by OEMs if T1 manufacturers comply the required specification.