Prognóstico de falhas utilizando Neo-Fuzzy-Neuron com aprendizado on-line

Abstract

Intelligent systems for fault prognostics in industries have brought important contributions in terms of safety and economy, making them indispensable, and motivating more and more research in the area. In prognostics problems, what is sought is to predict the Remaining Useful Life (RUL). In general, historical data and the technique of forecasting time series, of parameters indicative of degradation, multi-step ahead. In this context, this work addresses the use of a Neurofuzzy structure (Neo-Fuzzy-Neuron), with online learning, to estimate the RUL. It was proposed to use interval weights in the Neo-Fuzzy-Neuron network, to perform the prediction of RUL in an interval way, and thus obtain a RUL with a conservative and an optimistic value. The methodologies are applied in three databases, well known in the literature, to assess prognosis problems, they are: lithium ion batteries; wear of cutting tools on the Computer Numeric Control (CNC); and wear of bearings, database PRONOSTIA. The results obtained showed that the proposed methodology is very promising for the fault prognostics in processes/equipment.