Audio-based anomaly detection for industrial equipment has gained attention due to its low cost, non-intrusiveness, and ability to avoid interfering with normal operations. However, traditional audio-based detection methods often fail in scenarios involving domain shifts caused by changes in operating conditions and environmental factors, leading to frequent misjudgments. To address these challenges, this paper proposes a novel Unsupervised Adversarial Distillation Domain Adaptation (UADDA) method. This method reduces the distribution differences between source and target domains using domain adversarial adaptation while leveraging knowledge distillation to acquire structured knowledge from the source-domain teacher model. This enhances the robustness of feature representation and mitigates the interference of domain adversarial training on the main task, thereby improving anomaly detection accuracy in both source and target domains. Experimental evaluations on the MIMII DUE dataset demonstrate that, compared to mainstream methods, this approach achieves improvements of 3.59% in the source domain and 8.65% in the target domain, effectively enhancing the model’s detection robustness under different domain conditions.