Wi-Fi CSI-based Vibration Sensing for Industrial Fault Detection

In the modern landscape of ubiquitous wireless technologies, WiFi sensing has emerged as a promising method to gather environmental data through WiFi Channel State Information (CSI). This paper presents a comprehensive framework designed to identify frequency levels of vibrating objects and detect faults in industrial settings. Passive sensing using CSI data as in the Industrial Internet of Things (IIoT), wireless sensing has become increasingly popular for fault detection. The proposed methodology comprises four key stages: setting up the experimental environment, collecting CSI data, preprocessing the data, and predicting frequency levels and faults. CSI data were meticulously gathered across a frequency range of 0 to 20 Hz. Subsequent data processing involved feature selection using Recursive Feature Elimination with Cross-Validation (RFECV) and comparison with other feature selection techniques after preprocessing the raw CSI data. The processed data were then analyzed using both Machine Learning (ML) and Deep Learning (DL) models. Among the ML models, CatBoost demonstrated superior performance for frequency identification, achieving an accuracy of 95.52%. For fault detection, the Random Forest (RF) model excelled, attaining an accuracy of 98.77%. These models were evaluated based on various metrics, including accuracy, precision, recall, and f1 score, confirming their efficacy and robustness in industrial applications. The findings highlight the potential of Wi-Fi CSI-based sensing as a reliable and efficient solution for industrial fault detection and frequency identification.