Fault prognostic algorithm based on multivariate relevance vector machine and time series iterative prediction

To cope with prediction uncertainty and multiple fault features in fault prognostic application, an algorithm based on multivariate relevance vector machine (MRVM) is presented. Under certain preconditions, fault prognostics is firstly transformed to a time series prediction problem. Based on MRVM and matrix partitioning, it extends the existing time series iterative multi-step ahead prediction to the application with multiple fault features. Key variables in iteration are computed approximately by Monte Carlo sampling which effectively solves the problem that some integrals can't have analytical solutions and at the same time overcome the constraint of vector machine kernel selection in traditional algorithm. Prognostic outputs are of the form of probability distribution which is more suitable for prognostic application. Finally, A simulation experiment is adopted which demonstrates the effectiveness of the proposed algorithm.