Temporal Classification of Turbofan Engine Health using Elman Recurrent Network

Abstract

Prognosis and health management (PHM) plays an essential role in condition-based maintenance routines. For such purposes, academy and industry have devoted considerable efforts into providing efficient, safe and reliable solutions. In this regard, we aim at contributing to this field by proposing a temporal classifier for engine's health state identification based on the Elman recurrent neural network. The evaluation of the proposed approach involves a benchmarking data set originated from the C-MAPSS, a flexible turbofan engine simulation by NASA. A comprehensive performance comparison with state of the art approaches is then carried out. The proposed system is able to identify engine's total degradation 125 steps in advance, with 86.21% of confidence and low false negative rate, i.e. less than 2% of engines faulty conditions are identified as normal. With a temporal-based classification, the proposed approach reaches over 95% of accuracy on turbofan diagnosis.