Prototype-guided feature-space diffusion for zero-shot diagnosis of compound faults in rolling bearings
Year: 2026
Type: article
Abstract: Compound fault samples of rolling bearings are difficult to obtain in industrial rotating machinery because such faults occur infrequently and are costly to reproduce. This study proposes a prototype-guided feature-space diffusion method for zero-shot diagnosis of compound faults in rolling bearings. First, a multi-scale one-dimensional encoder extracts discriminative representations from four-channel vibration signals. Then, a structured fault attribute vector is constructed to describe fault components, compound cardinality, and location-related prior information. Conditioned on these attributes, a diffusion generator synthesizes pseudo-features for unseen compound fault classes in the learned vibration-feature space. To improve feature reliability, prototypes aggregated from semantically related seen classes are incorporated into reverse denoising to guide feature synthesis and reduce generation bias. The synthesized pseudo-features are used to train an unseen-class classifier. Experiments on the HDU-BFDS and Paderborn bearing datasets show that the proposed method improves zero-shot compound fault recognition under cross-condition and external-validation scenarios.
Authors: Yi Ren(Conceptualization, Funding acquisition, Methodology, Project administration, Resources, Supervision, Writing – review & editing), Chenxi Li(Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing), Fangjun Luan(Conceptualization, Funding acquisition, Project administration, Resources, Supervision), Shuai Yuan(Conceptualization, Funding acquisition, Project administration, Resources, Supervision), Ning Jin(Conceptualization, Funding acquisition, Project administration, Resources, Supervision)
Institution: Shenyang Jianzhu University
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: bronze
Awards: G35658185, G9001395836
OpenAlex ID: W7235651021
