A model for predicting the acceleration time histories induced on a two-wheeler mechanical frame has been set up and validated. The model is based on the Artificial Neural Network (ANN) technique and does not introduce any simplifying hypothesis for the physics governing the system.
The model accounts for different road roughness, variable vehicle speed and adjustable suspension pre-load, but other variables of interest can in principle be added without changing the model architecture. The height profile of the road is considered as a random, steady-state process that is characterized by its power spectral density (PSD). The functions describing the road PSD have been experimentally measured for the pavements of interest.
During this study, ANNs have proven to be a viable, economical way of modeling suspension behavior. Feedforward neural networks have been successfully trained by means of the backpropagation algorithm. Such networks provide accurate predictions when compared with experimental time histories recorded during typical riding conditions.