Beam choppers, as critical rotating components for regulating high-energy particle beams, are essential for maintaining the stable operation of large-scale scientific facilities through effective fault diagnosis. Conventional diagnostic methods often assume identical distributions between training and testing datasets and require abundant labeled samples to build robust models—assumptions that rarely hold in practical high-energy beam chopper scenarios. To address these limitations, we propose a digital twin-driven adversarial domain generalization with a diffusion model (DTADG) framework enhanced. First, a a denoising diffusion probabilistic model is trained exclusively on normal operating data to generate digital twin data. Then, these features are mixed and fed into a synthesize latent-space representations. By treating real measured data as the target domain and the generated twin data as the source domain, we develop an improved adversarial domain generalization strategy to test its generalization capability. Experimental evaluation on beam chopper fault diagnosis demonstrates that DTADG delivers superior diagnostic performance under distribution shift and limited data conditions, offering substantial benefits for fault detection in high-energy beam choppers.

Domain generalization,Diffusion model,Digital twin,Generalization capability

暂无