Predicting Tire Handling Performance Using Neural Network Models 2004-01-1574
Recent studies have shown that complex vehicle components such as shock absorbers, rubber bushings, and engine mounts can be accurately modeled by combining laboratory measurements with neural network technology. These nonlinear dynamic blackbox models (also known as Empirical Dynamics1 models) make it possible to predict nonlinear and hysteretic component behavior over wide ranges of amplitude and frequency. The models can handle realistic input waveforms as well as multiple inputs and multiple outputs.
These techniques have now been applied to rolling pneumatic tires, to enable high accuracy predictions of tire and vehicle handling behavior. Models that predict high amplitude force components (three forces and three moments) using up to four randomly-varying inputs (radial deflection, slip angle, and camber angle, and slip ratio) have been successfully generated, using data obtained from MTS Flat-Trac III tire test equipment. The accuracy of these models will be demonstrated using multiple analysis methods.
Advantages of this approach include: (1) models may be constructed without detailed knowledge of the tire structure or properties, so extensive expertise is not required, (2) the models can be generated very quickly using modern desktop personal computers, (3) the models consist of low order algebraic equations, so they execute quickly, (4) the blackbox nature of the models makes it possible to communicate behavioral properties without disclosing proprietary design details.