Error estimation of a neuro-fuzzy predictor for prognostic purpose

Prognostic is recognized as a key feature as the estimation of the remaining useful life of an equipment allows avoiding inopportune maintenance spending. However, it can be difficult to implement an efficient prognostic tool since the lack of knowledge on the behavior of an equipment can impede the development of classical dependability analysis. In this context, the general purpose of the work is to define a prognostic system for which any assumption on its structure is necessary: it starts from monitoring data and goes through provisional reliability and remaining useful life by characterizing the uncertainty following from the degradation process. Developments are founded on the use of the evolving eXtended Tagaki-Sugeno system as a neurofuzzy predictor. A method to estimate the probability distribution function of the predicted degradation signal is proposed. It enables to perform a priori reliability analysis. The approach is based on a recursive calculation procedure and is thereby well adapted to online applications.