Vehicle activity recognition using machine data

Modern vehicles are recording a lot of information about their dynamical states, and, often, vehicles are also equipped with telematics systems that communicate and administrate this information on a cloud infrastructure. In the area of commercial vehicles in particular, the usage variability is typically very high – possible applications and usage scenarios of such machines are manifold and often depend on special customer groups and their markets and regions of use. Thus, for the vehicle development process, it is of special interest to obtain precise knowledge and information about the actual use of a vehicle or machine, e.g., in order to define specifically tailored design targets and testing requirements. Moreover, information about what the vehicle is currently doing, i.e., its activity can be used online, on-board on the vehicle, e.g., for condition monitoring and predictive maintenance systems. In this contribution, we propose a data-driven vehicle activity recognition algorithm based on machine learning methods, that identifies the type of use (e.g., digging in the case of an excavator) with the help of machine or vehicle state information. We propose a multi-step procedure to derive the vehicle’s activity and apply our approach to a wheel-based excavator.