ArXiv Preprint
With the increasing availability of data for Prognostics and Health
Management (PHM), Deep Learning (DL) techniques are now the subject of
considerable attention for this application, often achieving more accurate
Remaining Useful Life (RUL) predictions. However, one of the major challenges
for DL techniques resides in the difficulty of obtaining large amounts of
labelled data on industrial systems. To overcome this lack of labelled data, an
emerging learning technique is considered in our work: Self-Supervised
Learning, a sub-category of unsupervised learning approaches. This paper aims
to investigate whether pre-training DL models in a self-supervised way on
unlabelled sensors data can be useful for RUL estimation with only Few-Shots
Learning, i.e. with scarce labelled data. In this research, a fatigue damage
prognostics problem is addressed, through the estimation of the RUL of aluminum
alloy panels (typical of aerospace structures) subject to fatigue cracks from
strain gauge data. Synthetic datasets composed of strain data are used allowing
to extensively investigate the influence of the dataset size on the predictive
performance. Results show that the self-supervised pre-trained models are able
to significantly outperform the non-pre-trained models in downstream RUL
prediction task, and with less computational expense, showing promising results
in prognostic tasks when only limited labelled data is available.
Anass Akrim, Christian Gogu, Rob Vingerhoeds, Michel Salaün
2023-01-20
