Yueyuan Fan

For fault diagnosis of rotating machinery under unseen working conditions, existing generative domain generalization methods commonly suffer from the entanglement between fault semantics and domain shifts, the lack of explicit constraints on the direction and strength of extrapolation, and the difficulty of quantitatively screening the reliability of pseudo samples. To address these issues, this paper proposes a Conformal Residual Score Extrapolation Diffusion (CRSED) method for domain generalization fault diagnosis. The method explicitly decomposes the noise prediction of the conditional diffusion model into a fault-shared score and a domain-residual score, disentangling cross-domain stable fault semantics from source-domain-related distribution shifts; a residual score extrapolation consistency constraint is then introduced so that the domain-residual branch maintains smooth and predictable responses beyond the source-domain boundary; on this basis, a semantic-reliable adaptive boundary extrapolation strategy for pseudo-domain generation is proposed and combined with a class-conditional conformal semantic boundary criterion to adaptively determine the safe extrapolation radius and screen reliable pseudo unseen-domain samples, enlarging the coverage of the source-domain distribution while preserving the semantic reliability of the generated samples. Experiments on public bearing datasets and a bearing dataset collected on a self-built fluid system simulation platform show that the average diagnostic accuracy of CRSED is superior to that of the compared methods. The average diagnostic accuracies on the JNU, HUST, and Fluid System Simulation Platform (FSSP) datasets are 94.52%, 95.63%, and 94.02%, respectively.

Standard data-driven fault diagnosis methods typically rely on the assumption of identical label spaces across source and target domains. However, the emergence of novel failure modes in the target domain invalidates this premise, presenting a complex open-set fault diagnosis (OSFD) challenge where traditional closed-set techniques fail to generalize. To bridge this gap, we propose a novel Adaptive Prototype Network. This framework operates through a progressive two-stage paradigm: a warm-up phase that leverages prototype learning and domain adversarial training to establish coarse-grained alignment, followed by a refinement phase that optimizes the model via adaptive clustering and dual-risk minimization. Specifically, we introduce a dynamic sample weighting mechanism guided by a learnable unknown detector to suppress negative transfer. Extensive experiments on two distinct datasets demonstrate that proposed method effectively aligns shared classes while robustly segregating unknown faults, outperforming state-of-the-art baselines.

Accurate remaining useful life prediction is pivotal for the reliability of lithium-ion battery management systems but is frequently challenged by the non-linear coupling of irreversible degradation trends and transient capacity regeneration phenomena. To address this, this paper proposes a novel hybrid framework specifically designed to integrate physics-informed signal decomposition with data-driven deep learning. First, we introduce an adaptive decomposition scheme—variational mode decomposition optimized by an improved triangulation topology aggregation optimizer. This physics-informed stage effectively decouples the raw capacity trajectory into two physically distinct components: a global degradation trend and local regeneration fluctuations. Subsequently, a dual-stream data-driven strategy is developed to characterize these components: a statistical-guided spectral attention long short-term memory (SGA-LSTM) is proposed to capture the long-term dependencies of the trend component, while a temporal convolutional network with ordered neurons LSTM models the high-frequency dynamics of the regeneration component. By reconstructing the predictions from these tailored sub-models, the proposed method achieves a holistic assessment of battery health. Comprehensive validations on NASA and center for advanced life cycle engineering benchmark datasets demonstrate that the framework controls relative errors within 5%, exhibiting superior accuracy and robustness compared to state-of-the-art approaches.