When bearings operate under variable conditions, they are affected by environmental noise and fluctuations in operating parameters. Existing bearing fault diagnosis methods relying on single-source signals struggle in these situations because such signals often fail to provide comprehensive and stable fault information. To address this issue, this paper proposes a multi-source information fusion network model (MSIFNM) based on an attention mechanism. The model′s multi-scale feature extraction module captures more detailed fault features, while the two-stage attention module enhances features that are less sensitive to changes in operating conditions. The multi-source information feature weighting module adaptively assigns weights to the features based on their sensitivity to faults. The feature fusion and classification output module further integrates these weighted features and delivers classification results through fully connected and softmax layers. To validate the effectiveness of the proposed MSIFNM model, bearing datasets under variable speed and load conditions were used. Experimental results demonstrate that the MSIFNM model significantly improves the accuracy, stability, and adaptability of bearing fault diagnosis under variable operating conditions.