Prognóstico de falhas baseado em redes neurais convolucionais: uma análise para estimar a vida útil remanescente de turbinas turbofan

Abstract

This work proposes a deep learning method for estimating the remaining useful life (RUL) of turbofan turbines. It presents a convolutional neural network (CNN) architecture. The proposed method is applied to a NASA data set for the RUL estimate, by varying the set of sensors and the form of turbine degradation in two models: linear RUL and piecewise linear RUL. Experimental results are compared with each other and with other methods found in the literature; the results show that the proposed piecewise linear RUL method exhibits similar performance found in other published works in terms of root mean squared error and a specific error metric used in competitions.