[5K2-OS-38a] OS-38

12:45 PM - 1:00 PM JST(3:45 AM - 4:00 AM UTC)

[5K2-OS-38a-04] Detection on Two Anomalies Sounds : Classical Machine Learning vs. Quantum Machine Learning

〇Takao Tomono¹, Kazuya Tsujimura² (1. Keio University, 2. Toppan Holdings)

Keywords:

quantum machine learning,anomaly detection,time series

We aim to detect anomalies in multiple devices in manufacturing. Conventional manufacturing equipment requires numerous sensors, with vibration sensors attached for each frequency, resulting in increased computational costs. Furthermore, there is also the issue of extensive wiring for each sensor. To address these challenges, this study first investigated whether two types of anomalies could be simultaneously detected using a single microphone. We compared the accurate model-building capabilities of classical kernels and quantum kernels integrated into one-class SVM. In our experiments, we detected abnormal sounds from conveyors and chain belts and examined the distance dependency. The results demonstrated that quantum methods outperform classical methods in anomaly detection. While classification using classical kernels showed performance degradation with increasing sensor distance, quantum kernels maintained high accuracy and F1 scores even at longer distances. The superiority of quantum kernels was confirmed even when the microphone orientation changed in three directions. Furthermore, by plotting the data in feature space, each individual anomaly appeared in a separate location, suggesting the possibility of distinguishing which anomaly it was