Feature extraction, recognition, and classification of acoustic emission waveform signal of coal rock sample under uniaxial compression

Abstract

In this study, based on Mel frequency cepstrum coefficient (MFCC) method, the AE signal characteristics of coal and rock samples were extracted, and the stress state criterion based on signal features was constructed. By integrating back propagation (BP) neural network for deep learning of signal characteristics, the recognition, classification, and prediction of coal and rock materials were realized. The results show that the MFCC could characterize the variation law of the original signal, with the sharp fluctuation of the amplitudes of both the AE signal and MFCC when the rock stress was near the peak value. Considering the ratio of sample stress to peak stress as the stress state, the correlation between MFCC and stress state was analyzed. The BP neural network exhibited a high accuracy rate for the signal characteristics represented by MFCC, achieving an accuracy of more than 95% with a fast recognition speed. Notably, the evaluation results of neural network model were stable and reliable. Therefore, MFCC can be used to extract the AE waveform signal characteristics and evaluate the stability of stress state for coal and rock materials. The recognition, classification, and prediction of high-precision results of the two types of waveform characteristics of coal and rock can be achieved through BP neural network.

Introduction

With the increase of mining depth, physical characteristics and mechanical behavior of coal and rock mass become more complex, which seriously impacts the stability control and safe and efficient mining of mine tunnel surrounding rock. Simultaneously, the complex and changeable environment of deep coal puts forward higher requirements for mechanical equipment for monitoring and early warning of mine dynamic disasters.

Noteworthy, high accuracy of coal and rock material identification in heading and mining face is the primary problem encountered during realization of the adaptive adjustment of equipment working condition and disaster monitoring and prediction. In mining engineering, acoustic emission (AE) and micro seismic detection techniques are used to collect dynamic signals of coal and rock failure in laboratory and field experiments. They offer the advantages such as data synchronization, multi-parameter collection, real-time continuity of dynamic monitoring, rapid measurement, massive data collection, noise filtering, event locating, etc. Recently, micro seismic monitoring technology has been extensively applied to quantitative monitoring and early warning of rockburst1, 2, 3, 4 risk and geological fault⁵ of coal and rock dynamic failure in deep mines. On the other hand, at the laboratory scale, AE technology is widely used for the systematic exploration of the correlation between signal parameters and physical and mechanical properties in the process of rock failure,^6,7 with the objective of verifying the feasibility and accuracy of this method through the visualization of signal parameters.⁸

During the application of AE testing technology, some experts have used the Hilbert–Huang transform method to extract the time–frequency characteristics of AE waveform signal and used it as the evaluation method of rock damage and failure to achieve a high recognition accuracy during non-destructive testing in the field of mining engineering.^9,10 Based on the analysis of stress changes and AE parameters during the process of coal and rock failure, a quantitative evaluation system has been constructed for coal and rock stability analysis by evaluating the dimension theory.11, 12, 13

Currently, the research on AE detection technology in the field of mining engineering mainly focuses on the collected parameter signals, including the amplitude, ringing count, energy, impact number, and event arrival time. The study on waveform signals requires the integration of feature extraction technology in speech recognition with the signal frequency conversion of communication principle. The common feature extraction methods include zero-crossing rate analysis,¹⁴ discrete wavelet transform,¹⁵ frequency-domain linear prediction cepstrum coefficient method,¹⁶ and Mel frequency cepstrum coefficient (MFCC) method.¹⁷ Compared with other methods, MFCC has the characteristics of high accuracy and stable recognition. At the same time, advantages of MFCC including high-fidelity representation and stable recognition also make it a potential candidate in the field of automatic speech recognition. For example, Ahmad et al. encouraged the use of MFCC and Mel Gaussian filter to calculate the characteristic parameters in order to define the audio signal characteristics.¹⁸ Cao et al. concluded that MFCC could reduce the error rate of signal characterization.¹⁹

With the development of computer science and technology in recent years, the deep learning of neural networks has gained extensive research attention on the global scale. For example, Peng et al. applied the improved back propagation (BP) neural network to classify the feature signals extracted from people and vehicles’ seismic signals and achieved accurate and effective results.²⁰ Liu integrated the weight segmentation technology to the traditional BP neural network algorithm to improve the filtering performance of the neural network, and applied the method to the recognition and classification of electroencephalogram signals.²¹ The parallel computing, learning ability, and distributed computing of probabilistic neural network enable it to quickly calculate the feature parameters extracted by MFCC.¹⁸ Li et al. proposed a one–dimensional (1D) convolutional neural network model, which could be used to identify and classify electrocardiogram signals.²²

Notably, in the field of mining engineering, the correlation between the signal characteristics obtained by MFCC method and the stress state of coal and rock materials has not been determined till date. Moreover, the research on the combination of strong correlation via MFCC and deep learning is scarce. Based on the study by Wang et al.¹⁷ on automatic speech recognition of AE waveform in uniaxial compression experiments of coal and rock samples, a neural network deep learning method based on MFCC feature extraction method was proposed in this study to characterize the waveform signal characteristics. Specifically, the AE waveform data of the uniaxial compression process obtained in the laboratory were used to investigate study the MFCC distribution of waveform signals of different materials under different stress states. Furthermore, the correlation model between the waveform characteristics extracted by MFCC method and the stress states of different materials was constructed. With the high goodness of fit characteristic parameters extracted by MFCC method as the data of BP neural network, the deep learning of the characteristic parameters of coal and rock materials by using the network model was implemented to successfully realize the identification and classification of coal and rock materials based on the short–time AE waveform characteristics. Importantly, the conception of utilizing a classic BP neural network to train and learn the MFCC feature extraction results provides a new method for the AE research on coal and rock material damage. It is a new idea for the adaptive adjustment of coal and rock identification technology in the working condition of mine equipment.