Failure prediction by means of advanced usage data analysis

This paper introduces and discusses an evaluation method of new data sources for reliability purposes. Through the continuously increasing amount of electronic control units (ECU) in capital goods, usage data may become available to improve reliability practices and maintenance processes. Such ECU-data sources may be captured and correlated in numerous ways and thus result in confusingly large matrices, which are impractical for classic statistical failure analyses. Consequently a method to condense such data based on physical knowledge prior to further analysis is introduced in this paper. A second method is to be applied on pre-condensed data in order to incorporate unknown effects and degradation mechanisms. Both methods combined yield a sharper image of observed field reliability, increase reliability knowledge on the product and consequently enable innovative maintenance policies. These methods consequently allow prediction of future failures and thus new reliability based maintenance themes for capital goods. Such maintenance themes can be aimed at lowering costs by increasing availability and hence improve customer trust. The path from the predicted failure probability to a decision helper tool is backed with economic considerations.