This study proposes a novel approach for bearing fault diagnosis based on Domain Adversarial Neural Networks (DANNs) and multi-scale feature extraction. Three feature extractors at scales 32, 64, and 96 are constructed to capture diverse feature representations. The domain adaptation techniques integrated into the model significantly enhance fault diagnosis accuracy compared to traditional Convolutional Neural Networks (CNNs). Specifically, the Conditional Adversarial Domain Adaptation (CADA) method achieves higher accuracy than DANN, while the Multi-Scale Domain Adversarial Neural Network (MSDANN) outperforms both, particularly on challenging tasks. The integration of different scale domain adaptation networks not only improves accuracy but also adds stability to the system, as each feature scale excels at specific tasks. The results demonstrate the effectiveness of domain adaptation techniques in addressing cross-domain fault diagnosis challenges and achieving higher accuracy and stability in multi-task scenarios.