The Fan Anomaly Detection Method Based on an Improved CBAM Attention Mechanism with MobileNetV3

In factories, fans is an important part in production. However, how to ensure the safe and stable operation of it is always a difficult problem. In order to solve this problem, this study presents a fan anomaly detection method based on an improved CBAM attention mechanism using the MobileNetV3-Small model. In this model, the conventional SE module is replaced with an optimized CBAM, which uses dilated convolutions to enhance the model's sensitivity to anomalous sounds. At the same time, the model incorporates CutMix and Focal Loss as the loss function, further improving the model's robustness to imbalanced data. Additionally, to strengthen the nonlinear representation capacity of the architecture, we implement CELU activation functions as substitutes for conventional ReLU units within residual modules. According to experimental results, we can find that the proposed model achieves a classification accuracy of 97% in a -6dB noise environment, which is an improvement of 1.07 percentage points over the original MobileNetV3 model, demonstrating the effectiveness of this approach in fan anomaly detection.