Prediction of remaining useful life under different conditions using accelerated life testing data

Prognostics methods model the degradation of system performance and predict remaining useful life using degradation data measured during service. However, obtaining degradation data from in-service systems in practice is either difficult or expensive. Therefore, accelerated life testing (ALT) is instead frequently performed for validating designs using considerably heavy loads. This work discusses the methods and effectiveness of utilizing ALT degradation data for the prognostics of a system. Depending on the degradation model and loading conditions, four different ways of utilizing ALT data for prognostics are discussed. A similar transformation method used in ALT is adopted to convert accelerated loading conditions to field loading conditions. To demonstrate the proposed approach, synthetic data are generated for crack growth under accelerated loading conditions; these data are used for training a neural network model or identifying model parameters in a particle filter. The applied example shows that the use of ALT data increases the accuracy of prognostics in the early stages in all four cases and compensates for the problem posed by data insufficiency through the proposed method.