A fault diagnosis method for unknown domain complex dynamic system based on contrast diffusion model

Abstract

A fault diagnosis method for complex dynamic systems in unknown domains based on a contrastive diffusion model is disclosed, belonging to the field of complex system state monitoring and diagnosis technology. The main steps include: constructing and preprocessing source and unseen domain datasets; building a contrastive diffusion diagnostic model composed of a fault sample encoder and a diffusion neural network; defining a conditional diffusion model framework containing endogenous disturbances, modeling operating condition differences as endogenous Gaussian noise; introducing a contrastive learning mechanism and defining a joint loss function composed of diffusion denoising loss and contrastive discrimination loss; performing end-to-end training of the model based on source domain data; and using the trained model to perform reverse denoising reconstruction and fault inference on unseen domain samples. This invention achieves high-precision fault identification under unknown operating conditions without requiring target domain data for training, significantly improving the model's robustness and generalization performance in complex system environments.

Images