Abstract
Today, car manufacturers are able to guarantee in service reliability of the mechanical parts of their cars. However, they are still working on their optimisation under the market pressure (weight, consumption). The just enough design requires a detailed knowledge of the cars use conditions. Therefore, car manufacturers perform customer surveys in which a lot of data is collected. Due to the limitations of the recording capacity, some information may be still missing. An appropriate analysis of the available signals may allow the reconstruction of the missing data.
This reconstruction has been achieved with neural networks for two types of data. The first one aims at the calculations of the input forces in the wheel from the measurement of accelerations. The second case treats of the identification of the weight of the car. In both cases, a simple non-linear model with two degrees of freedom has been used to determine the correct choices to be made for the structure of the networks and their inputs. The application of the identification of wheel forces on test track measurements is presented at the end of this article.