Continual learning for rotating machinery fault diagnosis with cross-domain environmental and operational variations
Diogo Risca, Afonso Lourenço, Goreti Marreiros
TL;DR
This work tackles fault diagnosis in rotating machinery under cross-domain environmental and operational variations by formulating a continual learning framework that grows capacity across domains through a shared feature generator and domain-specific classifiers. A boosting-inspired domain selection and restricted replay strategy enables forward and backward transfer while mitigating catastrophic forgetting, achieving high average domain accuracy around $88.96\%$ and forgetting as low as $2.70\times 10^{-3}$. The approach is validated on a multi-domain rolling bearing dataset using Markov Transition Field spectrograms, with ablations showing the impact of domain ordering, replay buffer size, and noise robustness. Practically, the method offers robust, adaptable fault diagnosis for predictive maintenance in non-stationary industrial environments, and it informs design patterns for continual learning in automotive, aerospace, and heavy machinery monitoring.
Abstract
Although numerous machine learning models exist to detect issues like rolling bearing strain and deformation, typically caused by improper mounting, overloading, or poor lubrication, these models often struggle to isolate faults from the noise of real-world operational and environmental variability. Conditions such as variable loads, high temperatures, stress, and rotational speeds can mask early signs of failure, making reliable detection challenging. To address these limitations, this work proposes a continual deep learning approach capable of learning across domains that share underlying structure over time. This approach goes beyond traditional accuracy metrics by addressing four second-order challenges: catastrophic forgetting (where new learning overwrites past knowledge), lack of plasticity (where models fail to adapt to new data), forward transfer (using past knowledge to improve future learning), and backward transfer (refining past knowledge with insights from new domains). The method comprises a feature generator and domain-specific classifiers, allowing capacity to grow as new domains emerge with minimal interference, while an experience replay mechanism selectively revisits prior domains to mitigate forgetting. Moreover, nonlinear dependencies across domains are exploited by prioritizing replay from those with the highest prior errors, refining models based on most informative past experiences. Experiments show high average domain accuracy (up to 88.96%), with forgetting measures as low as .0027 across non-stationary class-incremental environments.