Deep Learning for Improved System Remaining Life Prediction

Abstract

Prognosis of manufacturing system performance degradation is essential to operation safety and efficiency, and provides the basis for predictive maintenance scheduling. Complex physical mechanisms underlying machine operations often impose challenges to accurate health state assessment. This paper presents a data-driven approach to tracking system state degradation and consequently, predicting the remaining useful life, based on the Long Short-Term Memory (LSTM) network. Using aircraft engine fleet as an application context, the developed method reveals the temporal-dependency embedded in sensor data streams as the basis for engine degradation prediction. Good performance in engine remaining useful life prediction is demonstrated.