Answers are grounded in this paper and cite their sources.
Adaptive multimodel fusion under complex operating conditions and imbalanced data distributions remains a challenging problem in rotating machinery fault diagnosis, where model reliability varies significantly across samples and operating conditions, making fixed or heuristic fusion strategies suboptimal. To address this issue, this paper proposes a reinforcement learning-based autonomous multimodel weight allocation framework (RL-AMWF), which formulates multimodel fusion as a sample-dependent stochastic decision-making problem under uncertainty. Unlike conventional ensemble methods that rely on static or manually designed weighting rules, the proposed framework learns a policy that dynamically allocates model contributions according to uncertainty-aware observations, enabling adaptive fusion under varying diagnostic conditions. A unified state representation is constructed by integrating feature semantics, predictive distributions, and uncertainty cues, while a stochastic policy is learned to output adaptive fusion weights. A composite reward function is further designed to resolve the trade-off between overall diagnostic accuracy, weak-class sensitivity, and allocation stability, enabling non-decomposable optimization of fusion behavior. Extensive experiments on public bearing and gear datasets as well as an industrial simulation dataset demonstrate that the proposed method consistently outperforms several baseline methods in terms of overall accuracy, macro-averaged recall, and weak-class recall. Moreover, the proposed framework exhibits superior robustness under cross-condition and noise perturbation scenarios, validating the effectiveness of reinforcement learning for uncertainty-aware multimodel fusion.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: