Some large electrical equipment in industrial environments operates in variable conditions, resulting in significant differences in the distribution of collected fault data, which reduces the model’s cross-domain migration capability. Based on this, this paper proposes a cross-domain fault diagnosis method utilizing complementary weighted adversarial contrast domain adaptation. Firstly, a multi-scale expandable residual convolution block is designed to extract more detailed features without increasing the number of parameters. Secondly, it is proposed to combine the target domain prediction high confidence metric with the uncertainty of sample entropy, using both metrics in adversarial training to guide the model to focus on important features and improve the reliability of cross-domain migration. Finally, experiments show that the average accuracy of this paper’s model for cross-domain fault diagnosis on the PU dataset is 94.13%. This demonstrates the effectiveness of this paper’s model in cross-domain fault diagnosis and the reliability of cross-domain migration.