Improving Semi-Supervised Learning for Remaining Useful Lifetime Estimation Through Self-Supervision

RUL estimation plays a vital role in effectively scheduling maintenance operations. Unfortunately, it suffers from severe data imbalance where machines near their end of life is rare. Additionally, the produced by machine can only be labeled after failed. Both these points make using data-driven methods for difficult. Semi-Supervised Learning (SSL) incorporate unlabeled that did not yet fail into methods. Previous work on SSL evaluated approaches under unrealistic conditions failure was still available. Even so, moderate improvements were made. This paper defines more realistic evaluation and proposes novel approach based self-supervised pre-training. The method outperform two competing literature supervised baseline NASA Commercial Modular Aero-Propulsion System Simulation dataset.