Incremental Learning-Based Rolling Bearing Prognostics with Granular Uncertainty Quantification

Resumo

Resumo

Predictive maintenance plays a vital role in ensuring operational continuity and stands as a cornerstone of modern industry, where sensors and cyber-physical systems enable continuous equipment monitoring. Within this context, estimating the Remaining Useful Life (RUL) of machinery through physical, statistical, or computational intelligence models has become increasingly critical. Nevertheless, conventional approaches often encounter limitations, such as high data requirements, significant computational costs, and limited interpretability. Additionally, accurately estimating confidence intervals for RUL predictions remains a challenging task. To address these issues, this study proposes a novel methodology for RUL prediction in industrial systems by introducing an evolving granular model based on the principle of justifiable granularity - the AutoCloud predictor - capable of delivering interpretable predictions and quantifying uncertainty, even under conditions of data scarcity.

palavras-chave

Palavras-chave

remaining useful life predictionincremental learningjustifiable granularityconfidence intervalsdata scarcityinterpretable predictionpredictive maintenance

links

Links e arquivos