Integrated acoustic-vibration analysis: a dual-channel fusion prediction method is used for coal gangue identification in top coal caving

Abstract Top coal caving, a predominant technique for thick coal seam extraction, currently encounters significant challenges related to ‘over-mining’ and ‘under-mining’ during operational implementation. Current coal-gangue identification techniques exhibit notable limitations, while various sensor-based detection methods face challenges in practical deployment at top coal caving faces due to the extreme operational environment. To solve this problem, based on the two measurement channels of acoustic signals and vibration signals, this study proposes a dual-channel fusion prediction method. This method integrates the different time–frequency domain features of the two signals respectively through weighted fusion and feature dimension reduction at the feature level. At the decision level, it conducts decision fusion on the three baseline models through the improved analytic hierarchy process to improve the recognition performance of the model. Ultimately, the proposed prediction method achieved an accuracy of 96.25% and an Matthews correlation coefficient of 0.9253 on the test set. In addition, this study proposed preprocessing modules for two signals. The reconstructed signals were obtained by performing hierarchical denoising on the intrinsic modal functions after modal decomposition of the signals. The fusion prediction method proposed in this study presents a feasible scheme for accurately identifying coal gangue. This work not only provides a strategic pathway for advancing coal-gangue identification technologies, but also establishes a novel paradigm for multimodal pattern recognition through the innovative integration of acoustic-vibration characteristics.