<p>The performance of random forest (RF)-based fault diagnosis methods may be limited by insufficient labeled samples and varying data distributions across different working conditions. To address these challenges, a novel dual distribution adaptation random forest (DDARF)-based fault diagnosis method is proposed in this paper. First, a supervised balanced distribution adaptation (SBDA) method is introduced to reduce distribution discrepancies by adaptively leveraging the importance of both marginal and conditional distributions. Second, data from relevant working conditions are utilized to construct the initial RF-based fault diagnosis module. Finally, a new node distribution adaptation (NDA)-based method for updating node splitting thresholds is developed, which integrates the Gini index (GI) and maximum mean discrepancy (MMD) to evaluate the effectiveness of node splitting. The effectiveness of the proposed method is validated on a bearing dataset and a high-speed train simulation platform. Experimental results show that the proposed fault diagnosis method achieves superior performance compared to existing approaches. Overall, the proposed method offers an effective solution for diagnosing faults in traction motors under multiple operating conditions.</p>