Answers are grounded in this paper and cite their sources.
Data-driven health-state assessment often assumes that training and testing data follow similar feature distributions. However, different scenarios within the same health state may have different feature distributions. If some scenarios are absent from the training set, the resulting mismatch may reduce classification accuracy. From a symmetry-related perspective, the health-state label structure remains unchanged even when the associated feature distributions vary across scenarios. To address this problem, a Prototype Credibility-driven Self-Evolution Mechanism (PCSEM) is proposed. It uses multiple prototypes to represent each health state and evaluates their credibility based on accumulated matching evidence. Based on prototype states and new evidence, PCSEM updates existing prototypes, creates new ones when necessary, and merges redundant ones. The method is evaluated on the Case Western Reserve University (CWRU) bearing dataset, where only a subset of scenarios is available for training. Experimental results on the CWRU bearing dataset demonstrate the effectiveness of the proposed PCSEM, achieving classification accuracies ranging from 86.39% to 98.98% across eight experimental groups with relatively low active query ratios.
No takes yet. Share an insight, caveat, or question.
Shen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: