Predicting critical failures using physics of failures: opportunities and challenges

In this paper, a generic approach for applying physics of failure methods for component life prediction is described, which has been applied to several cases in the military domain in recent years. Three of these cases are discussed in more detail to demonstrate the potential and limitations: a CV90 vehicle sprocket wheel, a printed circuit board from a naval radar system and a shock absorber in a NH-90 helicopter. After that, the challenges encountered during development, testing and application of these models are discussed, and potential solutions and directions for research are indicated. The first challenge is the selection of the (critical) parts. The second challenge is the validation of the developed models, which suffers from the lack of well-documented failure data. The third challenge addressed is the link with data analytics. In recent years a lot of additional sensors have appeared in many weapon systems. However, interpretation of datasets and data analyses without proper knowledge on the system and its failure behaviour appeared to be rather difficult. Suggestions for combining artificial intelligence methods with physics of failure will be given, heading for the development of hybrid prediction methods.