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

Abstract

The evolution of manufacturing toward smart factories has underscored major challenges in equipment maintenance, particularly the dependence on numerous contact sensors for anomaly detection, leading to increased sensor complexity and computational costs. This study explores the use of quantum kernels to enhance anomaly detection based on noncontact sensors. We hypothesize that the expressive power of quantum feature spaces can effectively discriminate among multiple anomaly types using fewer sensors. Experiments were conducted on two types of manufacturing equipment a conveyor and a chain belt machine where a single directional microphone was placed at varying distances (0 to 3 m) to capture audio data. The signals were processed using autoregressive (AR) modeling to extract coefficient based features, which were then mapped into quantum feature space via quantum kernels for one class SVM classification. The quantum kernel classifiers achieved consistently high accuracy and F1 scores (more than 0.92) across all distances, while classical counterparts exhibited significant degradation beyond 0 m. Visualization of the feature space revealed clear separability, with distinct quadrants corresponding to different anomaly types: conveyor anomalies were mainly distributed in the second quadrant, and chain belt anomalies clustered in the fourth. These results demonstrate that quantum kernel methods can achieve robust, multi class anomaly detection in noisy factory environments using minimal non contact sensors. This represents a significant advancement toward quantum enhanced smart factories with reduced sensing infrastructure and improved maintenance efficiency. This work has been accepted for presentation at IEEE QCE25 and will appear in the IEEE Xplore Digital Library.