A Markov chains prognostics framework for complex degradation processes

This paper presents a prognostics methodology to deal with complex degradation processes for which condition monitoring data constitute the only available source of physical information. The proposed methodology is general, but here it is illustrated and tested using a case study about fatigue crack propagation in metallic structures. The prognostics method relies on a stochastic damage model based on Markov chains, which is embedded within a sequential state estimation framework for remaining useful life prediction and time-dependent reliability estimation. As key contribution, the resulting prediction and updating equations are shown to be obtained as closed-form expressions, whereby analytical equations for the remaining useful life and time-dependent reliability are obtained as by-product. Original contributions to the model parameter inference and the minimum required amount of data for prognostics are also provided.