Characterizing Rockbursts and Analysis on Hilbert-Huang Transform Spectrum of Microseismic Events, Shuangjiangkou Hydropower Station, Based on Microseismic Monitoring

Abstract

The Shuangjiangkou hydropower station in China has complex geological conditions with high in situ stress. During the tunnel excavation, rockbursts occurred frequently, which seriously affected construction progress. Microseismic (MS) monitoring technology was used to explore rock MS activities to predict rockbursts. The MS monitoring system can capture a large number of MS signals. Based on Hilbert–Huang transform (HHT) instantaneous frequency analysis technology, using MATLAB software (R2022a) to write a program to convert the MS waveform, the frequency and energy characteristics of MS signals at a certain time can be obtained. By analyzing the frequency and energy characteristics of every event, the microseism active areas can be determined, and then rockbursts can be predicted scientifically. This paper selected two different construction sites, which were the main powerhouse and the access tunnel in the main powerhouse, as the research background. Introducing HHT instantaneous time–frequency analysis technology conducted MS event dynamic analysis and predicted rockbursts. The HHT spectrum scientifically and comprehensively displayed MS signal frequency characteristics at a certain time and reflected the change laws of signal instantaneous energy and local abrupt change information. The results indicated that some parameter anomalies in the event spectrum can predict rockbursts. For complex tunnel construction conditions, the HHT time–frequency analysis technology can realize a new idea of using a single-channel signal to predict rockbursts, which was very meaningful.

1. Introduction

Rockburst is a common engineering disaster. In order to more accurately predict rockburst, so as to ensure smooth construction, researchers have used different methods and different angles to study the rockburst process and mechanism. The research methods for rockbursts include theoretical analysis, indoor tests, on-site tests, and numerical simulation. Representative rockburst theories include strength theory, energy theory, stiffness theory, and propensity theory. Researchers use theoretical analysis to study the rockburst mechanism and predict its occurrence. Indoor tests include uniaxial, biaxial, triaxial, and dynamic tests. On-site tests include on-site exploration, stress monitoring, microseismic monitoring, and so on. Researchers use different software to establish different models to analyze rockburst processes. Due to the limitations of experimental conditions and costs, there is relatively little analysis and summary of on-site rockbursts. However, on-site rockburst tests are the most practical and can better reflect the process of rockburst incubation and occurrence. The rockburst mechanism is complex, and rockburst prediction is difficult. Therefore, the prediction and prevention of rockbursts have always been a research hotspot. Naji et al. [1] used empirical approaches to evaluate the proneness of rockburst occurrence. Numerical simulation was used to predict the actual failure zone well. This method can effectively analyze the mechanism of rockburst, but it cannot grasp the internal activity status of surrounding rocks in real time and predict the occurrence of rockburst in a timely manner. Lee et al. [2] summarized the geotechnical issues encountered, including the various modes of rock mass instability and geological hazards, and introduced the current analytical techniques used for prediction in China. This article provided a comprehensive explanation of rock engineering technology. Rehman et al. [3] discussed the major challenges that were encountered during construction herein along with their countermeasures. The research results of this article have an important guiding significance for the excavation of deep-buried tunnels with complex geological environments. Zhang et al. [4] selected four typical rockburst events and described their temporal and spatial characteristics in detail. The research results were of great help in understanding the rockburst mechanism. They had relatively little research on the activity state inside surrounding rocks, which was very important for predicting rockbursts.

During the excavation process, the rock mass will inevitably produce different-scale microcracks, which contain rich information [5,6]. Microseismic monitoring technology is a three-dimensional spatial monitoring method that can form a spatial network structure by arranging a relatively small number of accelerometers, enabling real-time collection of microcrack information inside the rock mass. Then, by analyzing the microcrack information, the microseismic activity laws inside the rock mass can be better understood, revealing the precursor information of rock mass engineering instability and delineating the main damage zones and potential danger zones of the tunnel surrounding rock, which can provide a reference basis for later construction plans. The main characteristics of microseismic monitoring technology include: (1) It can directly determine important parameters such as the time, space, and energy release intensity of microcracks inside the rock masses; (2) Accelerometers can be deployed in areas far away from rock mass that are prone to damage, ensuring the safety of monitoring personnel and equipment, and ensuring that the monitoring system can operate for a long time without being damaged; (3) It can cover a large monitoring area with a limited number of sensors; (4) It can monitor in real time and transmit data remotely.

Microseismic monitoring equipment can capture rock fracture signals in real time. Through in-depth analysis of signal characteristics, we can obtain rich source information, which can be used as the basis for evaluating the stability of rock mass [7]. Compared with other waveform parameters, the frequency-domain characteristics are often unique [8] and relatively stable [9]. The spectrum distribution is an important parameter that can fully reflect the rock failure information carried by the acoustic emission waveform [10]. The MS signal frequency contains rich source information, and the change characteristics of different sequence frequency components can reflect different activity states inside surrounding rocks [11,12]. It has unique advantages and uses the frequency characteristics to conduct rockburst early warning and stability analysis for deep-buried tunnels. Some studies have shown that the high frequency of an MS event corresponds to a small-scale crack [13]. Therefore, the study on the change characteristics of the MS event frequency spectrum has great significance for rockburst early warning and obtaining the MS activity laws.

The spectrum of acoustic emission/MS can better reflect the characteristics of rock damage [14]. An increase in the low frequency of acoustic emission can indicate an imminent rockburst [15,16]. Some studies have shown that lower frequency or a trend of spectrum moving to the low-frequency band is rockburst precursor signals [17,18,19,20,21]. With the rapid development of signal processing technology, a lot of time–frequency analysis theories have been proposed. In recent years, the HHT has been widely used to process nonlinear signals. Huang et al. [22] first applied the HHT method to ocean analysis. Some scholars [23,24,25] applied the HHT to the spectrum research of seismic signals. Khaldi et al. [26] applied the HHT to speech signal denoising with different noise levels, and it was proved that the method was effective. Based on the HHT theory, Mao et al. [27] improved the empirical mode decomposition, improved the decomposition accuracy, and proved the effectiveness of the improved HHT method for low SNR nonstationary signal analysis. Duan [28] applied the HHT to image restoration for the first time and achieved good results. Yu et al. [29] introduced the HHT into gear fault diagnosis for the first time and proposed the local Hilbert energy spectrum concept. Good test results have been obtained. Lian [30] applied the HHT method to seismic phase analysis of seismic signals and achieved good results. Shi et al. [31] proposed a signal analysis method for acoustic nondestructive detection of full-length bonded bolts based on HHT. Yuan et al. [32] used the HHT method to analyze the blasting vibration signal in a lead-zinc mine in South China, and accurately identified the detonator initiation time of seven time delay levels. Veltcheva and Soares [33] applied the HHT method to identify different components of the spectra of real sea waves. Hamdi et al. [34] used the HHT for the extraction of a new relevant damage descriptor to be adopted for acoustic emission (AE) pattern recognition in order to help understand the damage process. Stroeer et al. [35] showed that complementing the HHT with techniques such as zero-phase filtering, kernel density estimation, and Fourier analysis allows it to be used effectively to detect and characterize signals with low signal-to-noise ratios. Litak et al. [36] analyzed the cutting force amplitude and frequency by means of wavelets and HHT. Numerous studies have shown that the HHT method is a relatively scientific signal processing method. Therefore, we can also introduce this method into rockburst prediction.

This paper focuses on rockburst prediction of the tunnels of the Shuangjiangkou hydropower station. The on-site construction environment is very complex, and fracture characteristics inside surrounding rocks are also complex. Therefore, rockbursts occurred frequently and were difficult to predict. The MS signals released contained rich information. The HHT spectrum scientifically and comprehensively displayed MS signal frequency characteristics and reflected the change laws of signal instantaneous energy and local abrupt change information. Compared with the conventional time–frequency analysis method, it can accurately capture the transient characteristics of signals. In this paper, HHT instantaneous frequency analysis technology was introduced. First, HHT instantaneous frequency analysis technology was applied to analyze and identify various on-site signals. Second, through rockburst analysis and prediction for two different sites, it was found that HHT instantaneous frequency analysis technology can achieve a good rockburst prediction effect, which had great significance for on-site construction and safety protection. The HHT time–frequency analysis technology by using a single-channel signal to predict rockbursts can adapt to complex tunnel construction environments.

2. HHT Theory and Signal Recognition

Based on HHT instantaneous frequency analysis technology, using MATLAB software to write a program to convert the MS waveform, the frequency characteristics and amplitude of MS signals at a certain time can be obtained. By analyzing the distribution and change characteristics of frequency and energy over time, the MS activity status inside surrounding rocks was analyzed to predict the possibility of rockburst occurrence.

2.1. HHT Theory

In 1998, Huang et al. [37] proposed a more effective nonlinear signal instantaneous time–frequency conversion technology, namely the Hilbert–Huang transform, which can identify the frequency characteristics of a signal at different time points, and provided a new idea for evaluating the stability of surrounding rock and other equipment failures. The algorithm uses empirical mode decomposition (EMD) signal filtering rules to decompose the signal into several characteristic components in the time domain, and then transform and combine their components. The transformation method can overcome the problem of insufficient signal frequency resolution.

2.1.1. EMD Principle and Algorithm [38]

First, find out all the maximum and minimum points on the time series X(t), and then use the cubic spline function to interpolate the maximum and minimum points of the signal, respectively, to obtain the upper envelope X_max(t) and lower envelope X_min(t) of a signal. The mean line m₁(t) obtained by connecting the upper and lower envelope lines in turn is as follows:

$$m_1(t)=[X_{\max }(t)+X_{\min }(t)]/2$$

(1)

The first component h₁(t) is as follows:

$$h_1(t)=X(t)-m_1(t)$$

(2)

In the subsequent sifting processes, h₁(t) can be treated only as a proto-IMF. Then,

$$h_{11}(t)=h_1(t)-m_{11}(t)$$

(3)

After repeated siftings in this manner, up to k times, h_1k(t) becomes an IMF; that is,

$$h_{1k}(t)=h_{1(k-1)}(t)-m_{1(k-1)}(t)$$

(4)

The test is defined as

$$SD={\displaystyle \sum _{t=0}^T\left|{\frac{\left|h_{1(k-1)}(t)-h_{1(k)}(t)\right|}{h_{1(k-1)}^2(t)}}^2\right|}$$

(5)

Huang et al. [24] have proved that when the SD value is between 0.2 and 0.3, it can not only ensure the linearity and stability of IMF, but can also make it have a clear physical significance, and the extraction effect of waveform frequency characteristics is relatively good.

When h_1k(t) meets the requirements of SD, the h_1k(t) is taken as the first-order IMF.

$$c_1(t)=h_{1k}(t)$$

(6)

The residual r₁(t) is as follows:

$$r_1(t)=X(t)-c_1(t)$$

(7)

Regard r₁(t) as the decomposed signal, and obtain all residuals r_i(t) after multiple operations. The result is

$$r_{i-1}(t)-c_i(t)=r_i(t),i=2,3,\cdot \cdot \cdot ,n$$

(8)

By summing up (7) and (8), we finally obtain

$$X(t)={\displaystyle \sum _{i=1}^n{c}_i}(t)+r_n(t)$$

(9)

2.1.2. The Hilbert Spectral Analysis

The Hilbert spectral analysis method [37] can more intuitively reflect the change characteristics of the instantaneous frequency of nonlinear signals over time. Firstly, every IMF component of the signal is subject to the Hilbert transform, and the corresponding analytic signal A[c_i(t)] is obtained as follows:

$$A[c_i(t)]=c_i(t)+j[\frac{1}{\pi }{\displaystyle {\int }_{-\infty }^{\infty }\frac{c_i(t)}{t-\tau }}d\tau ]={\alpha }_i(t)e^{j{\theta }_i(t)}$$

(10)

where the instantaneous amplitude and instantaneous phase are as follows:

$${\alpha }_i(t)=\sqrt{c_i^2(t)+H^2[c_i(t)]}$$

(11)

$${\theta }_i(t)=\arctan \frac{H[c_i(t)]}{c_i(t)}$$

(12)

where the instantaneous frequency is as follows:

$${\omega }_i(t)=\frac{d{\theta }_i(t)}{dt}$$

(13)

Then, the IMF component c_i(t) can be expressed as:

$$c_i(t)=\mathrm{Re}[{\alpha }_i(t)\exp (j2\pi {\displaystyle \int f_i}(t)dt)]$$

(14)

Show ${\alpha }_i(t)$ on the time–frequency plane, and the Hilbert spectrum of c_i(t) is obtained.

$$H_i(t,f)=\left\{\begin{array}{c}{\alpha }_i(t),\\ 0,\end{array}\right.\begin{array}{c}f=f_i(t)\\ f\ne f_i(t)\end{array}$$

(15)

We finally obtain the analysis signal of the Hilbert spectrum.

$$x(t)=\mathrm{Re}[{\displaystyle \sum _{i=1}^n{\alpha }_i(t)\exp (j2\pi {\displaystyle \int f_i}(t)dt)}]$$

(16)

2.2. Signal Recognition

A drilling and blasting method was adopted at tunnels of the Shuangjiangkou hydropower station, which made the construction noise more complex. The MS system installed can capture a lot of signals. Figure 1, Figure 2, Figure 3, Figure 4 and Figure 5 show some signals in the time domain and their HHT spectrum. The amplitude unit of the amplitude–time curve is set to V (Volt). The vertical coordinate of the HHT spectrogram represents the frequency and the color represents the amplitude. HHT spectrum maps have high temporal resolution and display a lot of effective information.

Figure 1 shows a signal monitored during the drilling hole. Four cyclic waveforms occurred within 800 ms. With the change of frequency over time, there were also four energy concentration areas, which indicated that there were four cyclic waveforms. The frequency distribution (FD) range was about 300–600 Hz. The noise frequency was not obvious in the figure, because the noise amplitude was small, and it was automatically shielded. Due to on-site construction, many drilling signals can be monitored.

An excavation blasting event is shown in Figure 2. The HHT spectrum showed that the rock failure process was extremely complex during excavation blasting. The FD range was about 100–1300 Hz, and the energy distribution of 400 Hz to 900 Hz was relatively concentrated. There was no obvious energy concentration area over time. The frequency points were discrete. Only a part of the blasting excavation signal was intercepted in the figure. There was a clear difference between the blasting signal HHT spectrum and other signals.

Figure 3 shows a signal when a truck whistled. The FD range of the event narrowed over time. The event energy was not obviously distributed below 400 Hz. There was also a small amount of information distribution in relatively high-frequency bands (800–900 Hz). The signal duration is relatively long. The frequency points were discrete.

Figure 4 shows a typical MS signal on 3 January 2018. The information of the MS signal was mainly distributed in 0.06–0.09 s, with a short duration of 0.03 s. The FD range was about 300–600 Hz, the FD had a downward trend over time, and there was an obvious energy concentration area in the figure. The frequency of energy concentration areas was relatively low.

Figure 5 shows the waveform of a rockburst occurred on 13 January 2018. The information of the rockburst signal was mainly distributed in 0.045–0.095 s, with a relatively long duration of 0.05 s. The FD range was about 100–600 Hz, which was relatively wide. In the low-frequency band, the signal information increased significantly; that is, the energy distribution increased significantly. The signal had obvious coda, and its FD range narrowed over time. For rockburst events, the amplitude was larger, the released energy was higher, and the signal component was more complex.

3. Case Analysis

The Shuangjiangkou hydropower station is an important hydropower project located in Sichuan Province, China, as shown in Figure 6. The maximum dam height is 314 m. The hydropower station has complex geological conditions. The diversion and power generation system is arranged on the left bank. The main rock type of tunnels is granite, and the rock uniaxial compressive strength is 60–70 MPa. The maximum principal stress is 37.82 MPa. The total length of the access tunnel in the main powerhouse is 1473.31 m, and cross-sectional dimensions are 11.24 m × 8.7 m (width × height). The powerhouse caverns are mainly buried horizontally at depths of 400 m to 640 m and vertically at depths of 320 m to 500 m. The excavation dimensions of the main powerhouse are 132.56 m in length, 29.30 m in width, and 63.00 m in height. The layout of the main powerhouse and the access tunnel in the main powerhouse is shown in Figure 7. Due to the comprehensive influence of high in situ stress and structural plane during the tunnel excavation, a series of serious damage problems occurred frequently. Among them, the occurrence of rockbursts is the most prominent problem, as shown in Figure 8.

Xue et al. [39] introduced the MS monitoring system of the Shuangjiangkou hydropower station in detail. The main powerhouse (early excavation) and the access tunnel in the main powerhouse are monitored by six accelerometers. At least four accelerometers must be triggered simultaneously to determine the source position. The Paladin data acquisition system has an acquisition frequency of 20 kHz.

3.1. Case Analysis of the Main Powerhouse

On 7 January 2019, two slight rockburst events were monitored at the main powerhouse. The on-site construction progress had slowed down. The HHT spectrum was used to analyze the MS activity laws of the whole rockburst process inside surrounding rocks. The time–frequency information of the rockburst precursor was extracted to provide a scientific basis for efficient construction.

Figure 9a shows an MS signal captured at 00:06:18 on 7 January 2019. The apparent stress value of the event was 16.186 KPa, which was relatively high. The energy released was 170 J. The Es/Ep was 2.72, which indicated that the rock damage type was mainly tensile damage. The FD range of the event was about 200–600 Hz and had a significant downward shift phenomenon over time. The frequency value was relatively low during 0.06–0.08 s. There was some energy concentration in the low-frequency band (200–300 Hz), which indicated there were relatively large-scale cracks inside the rock. The signal lasted approximately 0.08 s.

Figure 9b shows an MS signal captured at 00:08:31 on 7 January 2019. The apparent stress value of the event was 13.394 KPa, which was relatively high. The energy released was 8 J. The Es/Ep was 1.53, which indicated that the rock damage type was mainly tensile damage. The FD range of the event was about 300–700 Hz during 0.01–0.05 s, and was about 300–450 Hz during 0.05–0.065 s. The FD of the event had a downward shift phenomenon over time. The signal lasted approximately 0.055 s.

Figure 9c shows an MS signal captured at 00:12:54 on 7 January 2019. The apparent stress value of the event was 25.308 KPa, which was relatively high. The energy released was 263 J. The Es/Ep was 4.17, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 200–600 Hz during 0.025–0.07 s, and was about 100–300 Hz during 0.07–0.085 s. The FD of the event had a significant downward shift phenomenon over time. The FD range was obviously low, and there was some energy concentration in the low-frequency band. It indicated that the scale of the crack was relatively large. The signal lasted approximately 0.06 s.

Figure 9d shows an MS signal captured at 00:15:04 on 7 January 2019. The apparent stress value of the event was 19.756 KPa, which was relatively high. The energy released was 103 J. The Es/Ep was 3.98, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 300–500 Hz during 0.025–0.07 s, and was about 150–300 Hz during 0.07–0.1 s. The FD of the event had a significant downward shift phenomenon over time. The frequency points had good continuity, and the FD range narrowed. The energy distributions of the event were mainly concentrated within 200–300 Hz. The signal lasted approximately 0.075 s. The frequency spectrum characteristics of the event were typical precursory information of rockburst, indicating that rockburst was likely to occur.

Figure 9e shows a rockburst signal captured at 01:07:33 on 7 January 2019. The apparent stress value of the event was 66.482 KPa, which was high. The energy released was 3060 J. The Es/Ep was 5.67, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 150–400 Hz during 0.025–0.09 s. The frequency energy distribution of the signal had no downward trend over time. The overall FD range of the event was relatively low. The frequency points had good continuity, and the FD range was relatively narrow. The signal lasted approximately 0.065 s. There was a relatively loud “explosion” sound. The time interval between the occurrence of the rockburst event and the previous event was 52 min. It indicated that real-time monitoring had sufficient time to ensure safe construction on site.

Figure 9f shows an MS signal captured at 03:02:48 on 7 January 2019. The apparent stress value of the event was 17.421 KPa, which was relatively high. The energy released was 77 J. The Es/Ep was 5.16, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 200–600 Hz during 0.02–0.065 s, and was about 100–300 Hz during 0.065–0.085 s. The FD range had a downward shift phenomenon over time. The energy concentration was in the low-frequency band. It indicated that after a rockburst occurred, large-scale cracks continued to occur inside surrounding rocks. The signal lasted approximately 0.065 s.

Figure 9g shows an MS signal captured at 03:05:10 on 7 January 2019. The apparent stress value of the event was 11.448 KPa, which was relatively high. The energy released was 52 J. The Es/Ep was 3.45, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 300–600 Hz during 0.025–0.07 s, and was about 100–300 Hz during 0.07–0.085 s. The FD range had a significant downward shift phenomenon over time. There was some energy concentration in the low-frequency band. The signal lasted approximately 0.06 s.

Figure 9h shows an MS signal captured at 03:14:51 on 7 January 2019. The apparent stress value of the event was 5.914 KPa. The energy released was 2 J. The Es/Ep was 11.23, which indicated that the rock damage type was mainly shear damage. The FD range of the event was about 400–700 Hz during 0.025–0.06 s, and was about 100–300 Hz during 0.06–0.07 s. The FD range had a downward shift phenomenon over time. The signal lasted approximately 0.045 s.

Figure 9i shows an MS signal captured at 03:17:13 on 7 January 2019. The apparent stress value of the event was 7.224 KPa. The energy released was 10 J. The Es/Ep was 5.78, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 400–700 Hz during 0.015–0.06 s, and was about 100–300 Hz during 0.06–0.07 s. The FD range had a downward shift phenomenon over time. Some energy concentration was in the low-frequency band. The signal lasted approximately 0.055 s.

Figure 9j shows an MS signal captured at 03:22:21 on 7 January 2019. The apparent stress value of the event was 14.008 KPa, which was relatively high. The energy released was 136 J. The Es/Ep was 1.44, which indicated that the rock damage type was mainly tensile damage. The FD range of the event was about 200–500 Hz during 0.025–0.07 s, and was about 100–400 Hz during 0.07–0.085 s. The FD range had a downward shift phenomenon over time. The overall FD of the event was relatively low. The signal lasted approximately 0.06 s.

Figure 9k shows a slight rockburst signal captured at 03:25:17 on 7 January 2019. The apparent stress value of the event was 44.024 KPa, which was high. The energy released was 1172 J. The Es/Ep was 3.54, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 200–600 Hz during 0.02–0.06 s, and was about 100–300 Hz during 0.06–0.085 s. The FD range had a significant downward shift phenomenon over time. The FD range narrowed and the coda was developed. The signal lasted approximately 0.065 s. There was a relatively loud “explosion” sound.

Before a rockburst, for some events, the FD range had a downward shift phenomenon over time; some energy concentration was in the low-frequency band. The frequency points had good continuity. We need to analyze the HHT spectrum characteristics of every event promptly and in detail. If some events contain rockburst precursor information, it indicates an increased risk of rockburst.

3.2. Case Analysis of the Access Tunnel in the Main Powerhouse

On 13 January 2018, two rockburst events were monitored at the tunnel. There was a relatively loud “explosion” sound on site. The HHT instantaneous frequency analysis technology was used to analyze the rockburst process.

Figure 10a shows an MS signal captured at 07:05:26 on 10 January 2018. The apparent stress value of the event was 77.642 KPa, which was relatively high. The energy released was 154 J. The Es/Ep was 11.19, which indicated that the rock damage type was mainly shear damage. The FD range of the event was about 200–500 Hz during 0.02–0.11 s. The FD range had no downward shift phenomenon over time. The FD range was relatively stable. The energy distribution was relatively dispersed. The frequency points had good continuity. The signal lasted approximately 0.09 s.

Figure 10b shows an MS signal captured at 07:05:36 on 10 January 2018. The apparent stress value of the event was 30.660 KPa, which was relatively high. The energy released was 13 J. The Es/Ep was 6.84, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 300–450 Hz during 0.03–0.12 s. The FD range had no downward shift phenomenon over time. The frequency points had good continuity. The FD range of the event was relatively stable and relatively narrow. The coda was developed, and the energy distribution of the event was relatively dispersed. The signal lasted approximately 0.09 s.

Figure 10c shows an MS signal captured at 08:17:40 on 10 January 2018. The apparent stress value of the event was 59.019 KPa, which was relatively high. The energy released was 102 J. The Es/Ep was 8.06, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 200–600 Hz during 0.025–0.08 s. The frequency points had good continuity. The signal lasted approximately 0.055 s.

Figure 10d shows an MS signal captured at 22:20:58 on 11 January 2018. The apparent stress value of the event was 75.663 KPa, which was relatively high. The energy released was 261 J. The Es/Ep was 5.81, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 250–500 Hz during 0.03–0.09 s. The FD range had no downward shift phenomenon over time. The FD range of the event was relatively stable, and the coda was developed. The signal lasted approximately 0.06 s.

Figure 10e shows an MS signal captured at 02:14:37 on 12 January 2018. The apparent stress value of the event was 62.786 KPa, which was relatively high. The energy released was 79 J. The Es/Ep was 18.25, which indicated that the rock damage type was mainly shear damage. The FD range of the event was about 250–600 Hz during 0.03–0.1 s. The FD range had a downward shift phenomenon over time. The frequency points had good continuity. The FD range of the event narrowed. The coda was developed, and the energy distribution was relatively dispersed. The signal lasted approximately 0.07 s.

Figure 10f shows an MS signal captured at 02:13:49 on 13 January 2018. The apparent stress value of the event was 34.294 KPa, which was relatively high. The energy released was 167 J. The Es/Ep was 4.38, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 100–600 Hz during 0.03–0.13 s. The FD range had a downward shift phenomenon over time. The frequency points had good continuity. The FD range of the event gradually narrowed. The coda was developed, and the energy distribution was relatively dispersed. The signal lasted approximately 0.1 s.

Figure 10g shows an MS signal captured at 03:27:39 on 13 January 2018. The apparent stress value of the event was 24.959 KPa, which was relatively high. The energy released was 43 J. The Es/Ep was 7.74, which indicated that the rock damage type was mainly mixed damage. The FD range of the event was about 200–500 Hz during 0.025–0.13 s. The FD range had a downward shift phenomenon over time. The frequency points had good continuity. The FD range gradually narrowed. The coda was very developed, and the energy distribution of the event was relatively dispersed. The signal lasted approximately 0.105 s.

Figure 10h shows an MS signal captured at 03:38:41 on 13 January 2018. The apparent stress value of the event was 15.936 KPa. The energy released was 64 J. The Es/Ep was 2.06, which indicated that the rock damage type was mainly tensile damage. The FD range of the event was about 200–600 Hz during 0.025–0.07 s, and was about 300–400 Hz during 0.07–0.14 s. The FD range had a downward shift phenomenon over time. The frequency points had good continuity. The FD range gradually narrowed. The coda was very developed. The signal lasted approximately 0.115 s.

Figure 10i shows a slight rockburst signal captured at 04:07:44 on 13 January 2018. The apparent stress value of the event was 58.516 KPa. The energy released was 1154 J. The Es/Ep was 2.91, which indicated that the rock damage type was mainly tensile damage. The FD range of the event was about 100–700 Hz during 0.025–0.065 s. There was obvious energy distribution in the low-frequency band. The signal was relatively complex.

Figure 10j shows a rockburst signal captured at 04:19:06 on 13 January 2018. The apparent stress value of the event was 130.864 KPa. The energy released was 6597 J. The Es/Ep was 1.96, which indicated that the rock damage type was mainly tensile damage. The FD range of the event was about 100–700 Hz during 0.01–0.06 s. There was obvious energy distribution in the low-frequency band. The coda was very developed. The signal was relatively complex.

Before the occurrence of a rockburst, there were several indicators that needed to be noted. The event FD range had a significant downward shift phenomenon over time. The frequency points of some events had good continuity. The energy distribution of some events was in the low-frequency range. The FD range of some events gradually narrowed. Before the rockbursts, only some of these indicators may have occurred. For the comparison of multiple events, there was obvious energy distribution for some events in the low-frequency band over time. The HHT spectrum can well reflect the MS activity characteristics and timely feedback the anomalies inside the surrounding rocks on site.

4. Discussion

The rockburst mechanism is very complex, and it is also very difficult to predict accurately. Therefore, researchers need to explore scientifically by multiple methods and perspectives. Many researchers have found that the Hilbert spectrum can reflect the frequency characteristics and energy characteristics of a waveform at a certain time, and can more comprehensively display the characteristics of an MS waveform to feedback the MS instantaneous activity characteristics inside surrounding rocks.

Rock failure is a very complex process that contains a lot of information, so we need to constantly explore and identify. For MS signals or rockburst signals, due to their different energy released, damage types, and others, there are also some differences in their waveforms and frequency characteristics. Due to different factors inducing rockbursts, the precursor information of rockbursts is also different. According to the field monitoring, many MS events were captured before a rockburst occurred. These MS events contained different information. Some were typical MS events, and some events contained some rockburst precursor information. Therefore, we need to use certain technical means to analyze, identify, and accurately extract the precursor information of rockbursts to accurately predict the occurrence of rockbursts and even predict rockbursts in time and intensity.

The HHT spectrum scientifically and comprehensively displayed MS signal frequency characteristics at a certain time and reflected the change laws of signal instantaneous energy and local abrupt change information. For complex tunnel construction conditions, the HHT time–frequency analysis technology can realize a new idea of using a single-channel signal to predict rockbursts, which was very meaningful. The HHT instantaneous frequency analysis technology was introduced to analyze rockbursts that occurred at the main powerhouse and the access tunnel in the main powerhouse. It was found that before the rockbursts, the frequency of some events had a significant downward trend over time. From the HHT spectrum, the frequency points of some events had good continuity, the energy distribution of some events was in the low-frequency range, the FD range of some events gradually narrowed, and there was some other abrupt information. Rock failure information or rockburst precursor information was very complex. Therefore, rockburst precursor information may be partially or completely abnormal in frequency indicators. For some events, although the frequency has no obvious downward trend over time, the overall FD of events was relatively low. For the longitudinal comparison of multiple events, the energy distribution of some events increased in the relatively low-frequency band over time.

5. Conclusions

The MS system installed at tunnels of the Shuangjiangkou hydropower station can capture a lot of signals. Firstly, MS signal waveforms were recognized by using the HHT. Furthermore, introducing HHT instantaneous time–frequency analysis technology conducted MS event dynamic analysis and predicted rockbursts. The main conclusions are as follows:

(1)

For rockburst events, there was relatively high energy distribution in the low-frequency band, and the signal characteristics were relatively complex. If the energy released by an MS event was relatively high, there was obvious energy distribution for the event in the low-frequency band.

(2)

Before the occurrence of a rockburst, there were several indicators that needed to be noted. The event FD range had a significant downward shift phenomenon over time. The frequency points of some events had good continuity. The energy distribution of some events was in the low-frequency range. The FD range of some events gradually narrowed. Before the rockbursts, only some of these indicators may have occurred, or all of these indicators have occurred. For the comparison of multiple events, the energy distribution of some events increased in a relatively low-frequency band over time.

The HHT spectrum can well reflect the MS activity characteristics and timely feedback the anomalies inside the surrounding rocks on site. In this paper, the HHT instantaneous frequency analysis technology was introduced to reveal the instantaneous frequency change laws of the single-channel signal before rockbursts. A new idea of using a single-channel signal to predict rockbursts was realized.