Fatigue life prediction using multivariate gaussian process

A hybrid prognosis model is being developed for real-time residual useful life estimation of metallic aircraft structural components. The prognosis framework combines information from off-line physics-based, off-line data driven and on-line system identification based predictive models. The present paper focuses on the later two components of an integrated, hybrid prognosis model. These components are explicitly based on Gaussian process based data driven approach within a Bayesian framework. Fatigue crack behavior of Aluminum 2024 compacttension (CT) specimens under variable loading has been modeled using this multivariate Gaussian process technique. The Gaussian process model projects the input space to an output space by probabilistically inferring the underlying non-linear function relating input and output. For the off-line prediction the input space of the model is trained with parameters that affect fatigue crack growth, such as number of fatigue cycles, minimum load, maximum load, and load ratio. For the case of online prediction, the model input space is trained using features found from piezoelectric sensor signals rather than training the input space with loading parameters, which are difficult to measure in a real flight-worthy structure. In both the off-line and on-line case the output space is trained with known associated crack lengths. Once the Gaussian process model is trained, a new output space for which the corresponding crack length or damage state is not known is predicted using the trained Gaussian process model. Concepts are validated through several numerical examples.