Clustering analysis and recognition of acoustic emission signals as failure precursors in coal samples

Real-time monitoring and precursor identification of coal failure are crucial for providing early warnings of coal and rock dynamic disasters in coal mines. In this study, the acoustic emission (AE) response characteristics and crack evolution of coal failure precursors were analyzed under laboratory conditions. The AE signals of coal failure were classified using a two-step clustering method. Precursory AE signals were extracted and analyzed, followed by their identification using a support vector machine (SVM) algorithm. The results showed that Type V AE signals first appeared near the peak stress and exhibited distinctive features: low amplitude (hundreds of mV), long duration (hundreds of thousands of μs), large counts (tens of thousands), and high energy (tens of thousands of mV * ms). Type V AE signals were selected as the precursory signals of coal failure. An SVM model was subsequently trained to learn and identify these signals, achieving an average precision of 95.83%. The times when the SVM model first recognized the failure precursor AE signals of coal samples 1 and 2 were 19.442 s and 11.5686 s earlier than their peak failure times, respectively. The research results provide a valuable reference for using AE to identify coal failure precursors.