Quantum Kernel Anomaly Detection Using AR-Derived Features from Non-Contact Acoustic Monitoring for Smart Manufacturing

The evolution of manufacturing toward Smart Factories has highlighted critical challenges in equipment maintenance, particularly the reliance on numerous contact sensors for anomaly detection, resulting in escalating sensor and computational costs. This study investigates the application of quantum kernels to enhance anomaly detection using non-contact sensors. We hypothesized that quantum computing's expressive power could effectively discriminate among multiple anomaly types using fewer sensors. Our experimental setup involved detecting and classifying anomalies from two distinct manufacturing equipment: a conveyor and a chain belt machine using a single directional microphone positioned at varying distances ($0-3 ~\mathrm{m}$). Audio data was processed through Autoregressive (AR) models to extract coefficient features, which were then mapped into quantum feature space using quantum kernels for one-class SVM classification. Results demonstrated that quantum kernel implementations maintained near-perfect accuracy and $\mathbf{F} 1$-scores (consistently $\mathbf{> 0. 9 2}$) across all distances, while classical approaches showed significant performance degradation beyond the 0 m position. Feature space visualization revealed that quantum kernels effectively separated different anomaly types into distinct quadrants within a two-dimensional representation, enabling not only detection but also classification of multiple equipment failures. Specifically, under the third and fourth features space, conveyor anomalies consistently appeared in the second quadrant, while chain belt anomalies clustered in the fourth quadrant. This study demonstrates that quantum kernel methods enable significant anomaly detection in noisy factory environments using fewer non-contact sensors, representing an important step toward realizing quantum-enhanced smart factories with reduced infrastructure requirements and improved maintenance efficiency.