This content is not included in your SAE MOBILUS subscription, or you are not logged in.
Data Based Damage Prediction of Commercial Vehicles Using Bayesian Networks
ISSN: 1946-391X, e-ISSN: 1946-3928
Published October 07, 2008 by SAE International in United States
Citation: Lorenz, A. and Kozek, M., "Data Based Damage Prediction of Commercial Vehicles Using Bayesian Networks," SAE Int. J. Commer. Veh. 1(1):373-382, 2009, https://doi.org/10.4271/2008-01-2659.
For the estimation of life expectancy and dynamic fatigue of a machine, the overall load configuration of the typical application is of major importance. Regarding commercial vehicles, the load spectrum differs with the variation of machine parameters which requires costly measurements for analysis of damage. This article presents robust methods for the computation of characteristic values for the damage to a certain component. The methods are based on a hypermodel, which represents the relation between different machine configurations and the resulting characteristic values. Therefore, fewer typical machine configurations have to be measured. The statistical models of load and damage are made using the Rainflow counting algorithm and an extended version of Miner's Law. After the condensation into characteristic damage values, hypermodels for the relationship between these scalar values and the machine parameters are developed using Neural Networks. Due to the small amount of measurements and the high variance in the data, Bayesian regression is used to optimize performance and robustness. The method is validated using data from simulations and also tested on measured data of earth moving vehicles. Experimental results show that Bayesian techniques lead to a good prediction of the damage if the data quality is reliable, and they point out parameter configurations that should be measured additionally (Design of Experiment). Using the presented approach, the development procedure of commercial vehicles can be considerably economized by a drastic reduction of both time and measurement cost.