Prediction of Remaining Useful Lifetime of Membrane Using Machine Learning

Abstract

We present a novel analytical procedure estimating the remaining useful life (RUL) of complex systems or facilities based on degradation data obtained over time; we consider the maintenance characteristics of units that are incompletely repaired. We develop an extended prognostic model that accurately predicts the RUL; we use machine-learning featuring smoothing, logging, variable transformation and clustering to this end. The performance of a general model was more predictable than that of an extended model. A linear regression (LR) method was superior in terms of root mean square error prediction and an artificial neural network (ANN) was superior in terms of prognostics and health management (PHM) scoring. The procedure is both practical and efficient, and can be deployed in various industries, yielding low-cost prognostics even in low-expertise domains. The procedure can be applied to high-risk industries, aiding management decision-making in terms of the establishment of optimal, preventative maintenance policies.