Potential of multi-anomalies detection using quantum machine learning

Maintenance of production equipment is critical in manufacturing. Typically, machine learning models are trained on sensor data closely attached to equipment. However, as the number of machines increases, computational cost grows rapidly. In practice, anomalies are often identified by human operators through auditory perception, relying heavily on experience and intuition. In vibration analysis, especially, AR model coefficients combined with one-class SVMs are used for detecting anomalies. In this work, we explore the effect of substituting the classical kernel in the one-class SVM with a quantum kernel. Two experimental setups were used. The first involved a miniature racing car track, where the car passes over a patch of hook-and-loop fastener to generate abnormal sounds, which are recorded using a microphone. The second involved an open-belt drive, where chopsticks are inserted at specific times to produce crushing sounds, simulating sudden anomalies. Our results show a clear advantage of quantum kernels over classical Gaussian (RBF) kernels. On the miniature car track dataset, the quantum kernel achieved an accuracy and F1-score of 0.82, compared to 0.64 and 0.39 respectively for the RBF kernel. For the crushing device, the quantum kernel achieved perfect accuracy and F1-score (1.00), while the RBF kernel reached only 0.64 accuracy and 0.43 F1-score. These findings suggest that quantum kernels enhance the classification accuracy for diverse types of abnormal sound patterns, including both periodic and impulsive anomalies.