Feature Extraction and Unsupervised Classification of Roadway Fracture Signals: A Full-Section Wi-Fi Wireless Monitoring Approach

Aiming to address the challenge of the high-precision monitoring of underground coal and rock fractures, this paper proposes and verifies a roadway full-section synchronous monitoring method utilizing a Wi-Fi wireless sensor network. To address the inherent difficulties of detecting complex rock mass fractures through surface sensors, our methodology employs a synchronized array of surface-mounted vibration sensors covering key mechanical structural points. The feasibility of this approach is technically substantiated through the strict implementation of rigid coupling techniques—utilizing industrial-grade epoxy resin and customized metal mechanical fixtures—combined with hardware low-pass filtering to eliminate air gap attenuation and maximize the signal-to-noise ratio. Using this validated setup, we successfully extracted and manually verified 63 high-fidelity rupture events. The data reliability is further demonstrated through a comprehensive Python-based processing pipeline that calculates 17-dimensional time–frequency characteristics. Statistical analysis confirms that the extracted data strictly conforms to the physical laws of rock fracture, evidenced by a significant negative correlation between maximum amplitude and dominant frequency (r = −0.84, p < 0.001). Unsupervised clustering of these signals reveals excellent inter-class separability. By transparently substantiating the data acquisition and verification process, this study provides a publicly shared pilot dataset and methodology for algorithm evaluation and preliminary dynamic disaster mechanism exploration.