2025 International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD 2025)

Industrial rotating machinery often cannot collect all possible fault data at once. This paper presents an open-set anomaly detection network, TinyCNN-SE (Tiny Convolutional Neural Network(CN) with Squeeze-and-Excitation blocks(SE)), which appends a Cross-Entropy Softmax layer after SE-ResBlocks. Experiments on the Automobile Transmission (AT) Dataset show that when only one unknown fault class is present, the proposed method achieves over 90 % accuracy, and when multiple unknown classes appear, it still maintains over 80 % accuracy.
 

CNN, SE-ResBlocks, Open-Set Bearing Anomaly Detection