The response characteristics of Mel cepstrum coefficient of acoustic emission and intelligent identification of damage state of loaded coal-rock-coal structure

The stability of coal-rock-coal composite structure (CRCS) is crucial for the safe mining of coal mine. Accurate identification of damage state is of great significance for judging instability. Therefore, the uniaxial compression test of CRCS with different dip angles was carried out, the damage evolution law and mechanical properties was analysed. The response characteristics of Mel frequency cepstral coefficients (MFCC) of acoustic emission (AE) are analysed and the damage state of CRCS is recognised by the MFCC-Long short term memory (LSTM)-Autoencoder model. The results show that when one dip angle of CRCS is a certain value, with the increase of another dip angle, the mean value of uniaxial compressive strength (UCS) and elastic modulus (EM) increase first and then decrease. Some MFCC can accurately describe the damage state of CRCS. When after the 80% of UCS, there will be a noticeable increase or decrease. The LSTM-Autoencoder model can learn the characteristics of MFCC to identify the damage state of the CRCS. By the five-fold cross-validation, it can be seen that the average values of the four evaluation indexes of the MFCC-2 data reached more than 85%, which indicates that the stability and generalisation effect of the model are good.