International Conference
| Audio Data-driven Anomaly Detection for Induction Motor Based on Generative Adversarial Networks | |
|---|---|
| Year | 2022 |
| Month | October |
| Journal | 2022 IEEE Energy Conversion Congress and Exposition (ECCE) |
| Author | Jaehoon Shim; Taesuk Joung; Sangwon Lee; Jung-Ik Ha |
| Link |
http://ieeexplore.ieee.org/document/9947652
120회 연결
|
This paper presents an anomaly detection model for an induction motor. The proposed one uses a Generative Adversarial Networks(GANs) model, which receives audio data from multiple microphones as its input. The audio signals from each microphone are transformed into mel-spectrograms used as the model's input. The anomaly cases considered in this paper are bearing faults that occurred at the front and rear bearings. One thousand train data were obtained from the normal target motor running at various operating points. The 200 test data were selected from data obtained from three target motors (i.e., normal, front bearing fault, rear bearing fault conditions) at different operating points. The proposed method is a GANomaly[1]-based approach, and contextual loss is defined as an anomaly score. As a result, 0.9822 Area Under the Receiver Operating Characteristic (AUROC) and 92.52 % accuracy were achieved for the untrained anomaly data of front and rear bearing faults. The model training was performed with PyTorch in a Python environment. |
|
Seoul National University, 1, Gwanak-ro, Gwanak-gu,
Seoul, Republic of Korea, 08826 Bldg. 301, Rm. 606