Real-time monitoring of the operational status of mechanical equipment is crucial in industrial production. The scarcity of fault data, the predominance of unlabeled monitoring data, and the complex and variable operational conditions have gradually become prominent, which limit the accuracy of diagnostic models and affect their effectiveness in new operational conditions. To address these issues, a domain adaptation fault diagnosis method based on a multi-scale feature adaptive ConvNeXt is proposed in this paper. By integrating multi-scale feature fusion, multi-spectral attention mechanisms, and feature adaptive selection modules, the method enhances the ConvNeXt network’s capability to capture fault characteristics. It combines conditional domain adversarial techniques to align the cross-domain feature marginal and conditional distributions. Comparative experiments conducted on two bearing datasets demonstrate the effectiveness of the proposed method in addressing the challenge of sparse labeling under target operating conditions, enabling accurate assessments of the health status of mechanical equipment in target conditions.