Dynamic recognition of coal-rock interface based on hardness characteristic preference and multisensor information fusion

A multisensor information fusion method for coal-rock interface recognition based on hardness characteristic preference was proposed to overcome the impact of hardness differences between coal and rock on the accuracy of coal-rock interface recognition during the mining process. First, a coal-rock cutting experimental platform was established using 15 coal-rock specimens with three hardness characteristics (soft coal-hard rock, coal-rock with similar hardness, and hard coal-soft rock) and five coal-rock ratios. Then, cutting experiments were conducted, and frequency-domain analysis coupled with wavelet packet reconstruction was employed to construct a multicutting signal characteristic value database encompassing the current and triaxial vibration signals. Subsequently, membership functions for multicutting characteristic signals were developed based on minimum fuzziness principles under varying hardness conditions, with membership degree thresholds optimized via a particle swarm optimization (PSO) algorithm. Finally, a coal-rock interface recognition decision model was constructed by integrating the Dempster–Shafer (D-S) evidence theory with a multi-PSO framework. The experimental results demonstrate that the proposed method achieves a maximum recognition accuracy of 0.9626 for specimens with diverse hardness characteristics, reduces the uncertainty probability by up to 109.1 %, and yields a total error of 2.38 % (55.93 % reduction) in coal residue and rock intrusion scenarios. The approach provides a robust theoretical foundation and technical framework for advancing intelligent coal-mining systems.