Métodos de prognóstico híbrido baseados em filtro de partículas aplicados em uma caixa de engrenagens

Abstract

In the present work, two hybrid failure prognostics approaches based on particle fiters (PF) were developed to predicting Remaning Useful Life (RUL) of a gearbox. The gearbox is composed by a spur gear pair on which pinion has a tooth root crack. The first approach is the PF with artificial dynamic on parameters, on which parameters are treated as states and the FP is applyed to, so called, extended space. The second approach, it is the Particle Metropolis-Hastings (PMH), which unifies the PF with Markov Chain Monte Carlo (MCMC). First, the degradation and measurement model are implemented. The degradation model is based on the Paris Law, which describes the dynamic bahavior of crack propagation based on Stress Intensity Factor (SIF) and material parameters. In this work, the SIF is obtained by finite elements model with gear model. The measurement model developed relates the crack length with the Root Mean Square (RMS) variation and Kurtosis variation indexes extracted from the vibration signal. Three uncertainty sources are considered on these models: degradation model uncertainty, material parameters uncertainty, and measurements errors. With the models, the two prognostics approaches were tested and both prognostics approaches obtained success at estimating the crack length and the RUL of the gearbox. In addition, the two approaches estimate the material parameters.