Hybrid neural network model for history-dependent automotive shock absorbers

The method of numerical multi-body simulation is an often used and well-understood development tool in the automotive industry. In order to reproduce the ride comfort or handling behaviour of vehicles, mathematical models have to be built up. To achieve accurate simulation results, highly detailed component models are required. However, the formulation of appropriate physically-based model equations of complex automotive components (e.g. air springs, shock absorbers, rubber bearings, tyres, etc.) can be very difficult. To handle this, empirical modelling methods have been developed. Simple algebraic equations are used to describe complex system behaviour. This simplification is very effective, although it largely ignores the natural laws of mechanics and thermodynamics but is still capable to predict the component response. This article illustrates how to take advantage of this approach in numerical simulations. We describe the development of a hybrid automotive shock absorber model based on both spline and neural network (NN) approaches. By combining these different approaches, an accurate model is achieved without loss of variability. Non-isothermal laboratory force-displacement measurements of an automotive shock absorber are being used to estimate the parameters of the NN. As shown, the model replicates the measured data with sufficient accuracy, especially the hysteresis. Finally, we present a set of quarter-car simulations with a built-in hybrid NN shock absorber.