Passive Safety in Autonomous Vehicles: Machine Learning and Artificial Intelligence Applied to Human Body Model Positioning

The development of automated driving systems is raising new challenges for passengers safety. One of the reasons is that the driver and the passengers can engage in a wider range of activities. Thus, their posture varies more from what is nowadays considered to be conventional. Being less expensive, finite element methods are used to investigate new crash scenarios, including those in which passengers are sitting in non-conventional positions. However, positioning a human body model in a non-conventional posture is a very demanding process. The combination of machine learning and artificial intelligence allows to reduce the order of a model, thus drastically lowering both time and computing effort. This paper aims at making a reduced-order model of the THUMS. With the intention of doing so, four reduced-order sub-models are combined into one model. Each sub-model controls a limb and requires a specific database. Testing is required on the sub-models and the merged model. Both the database and the testing samples are computed using full finite element simulations. The reduced-order model should eventually allow human body model positioning in a few minutes, with a precision nearly as close as a full finite element simulation.