*4.1. Gold Mine Cook 4 in South Africa*

Performing in situ AE monitoring inside the deep mines in South Africa is challenging, because these mines are active working mines that come with an especially harsh environment. What is beneficial for AE monitoring is the hard rock environment that allows to observe seismic waves with frequencies far above 25 kHz from significant distances from more than 100 m [36]. For SHM, in situ AE monitoring was shown to be a valuable tool, because damage zones prone to failure can be identified beforehand and rock bursts can be analyzed in great detail after.

The SATREPS project [33], that took place in Cooke 4 Mine, near Westonaria in South Africa at about 1 km depth demonstrated how to monitor a mining front in quartzite from afar, using a network of 24 field AE sensors and 6 triaxial accelerometers installed in development tunnels approximately 20 m to 50 m below the stope (network dimension 95 m × 50 m × 30 m, monitoring area 100 m × 180 m × 50 m). Figure 13 shows configuration of the in situ AE monitoring system installed 1 km beneath the ground in the Cooke 4 shaft. This figure shows the top view of the stope and the positions of sensors. The black squares denote the positions of the borehole sensors. The hatched area represents the excavated area. The mining faces are typically 30 m across and 1 to 2 m high, and the stope is sub-horizontal orientated. By daily blasting, the mining face advances to the north by about 10 m per month (red arrows). The mining face was located about 20 m above the AE network.

**Figure 13.** Top view of the stope and the positions of the borehole sensors (black squares). The shaded area represents the excavated area. The mining faces are typically 30 m across and 1 to 2 m high and the stope is sub-horizontal orientated. By daily blasting, the mining face advances to the north by about 10 m per month (red arrows) [34].

Over a time period from 11 July 2011 to 24 August 2011 (approximately 50 days), about 290,000 AE events were recorded in trigger mode recording and localized using the joint hypocenter determination (JHD) approach and double-difference re-localization. The moment magnitudes MW of events ranged from −3.7 to 1.0 [53]. Figure 14 shows the JHD result of 289,015 events, that had RMS residuals smaller than 1.0 ms in a top view (a) at an elevation of 690 m above sea level (about 1 km beneath ground), and (b) NNE–SSW vertical projection viewed from the west.

**Figure 14.** Source locations of AE events determined by JHD in a top view (**a**) at an elevation of 690 m above sea level (about 1 km beneath ground), and (**b**) NNE–SSW vertical projection viewed from the west. Black squares indicate AE sensor locations. Thick cyan and red lines represent positions of the mining face on 21 July and 24 August 2011, respectively. Blue lines are outlines of the mining infrastructure. Box B outlines the area that is shown in Figure 15 in detail. A total of 289,015 AE events were located, with their dates of occurrence indicated by the color scale [34].

Black squares indicate AE sensor locations. Thick cyan and red lines represent positions of the mining face on 21st July and 24th August 2011 respectively. Blue lines are outlines of the mining infrastructure. The 289,015 AE events were located, with their dates of occurrence indicated by the color scale. Nearly all (90%) of the AE events occurred in a zone extending about 20 m ahead (north) of the active mining face, forming what is referred to as the stope-front cloud. AE events clearly clustered on structures' outlining planes of localized damage, which is why the authors referred to the monitoring as an advanced AE mapping technique.

Naoi et al. [84–86] relocated clustered AE events by using the double difference technique [94]. After a pre-selection of events, the cross-correlation technique for all event pairs whose inter-event distance was smaller than 4 m was applied. The number of events in Box B in Figure 14a before relocation was about 10,688. The number of events successfully relocated as AE events was 10,337. Figure 15a,b show relocated AE events in projections onto the x-y plane and x-z plane, respectively. A local coordinate system aligned with the clear discernible plane is used. The x coordinate is along the strike direction (positive eastward), the y coordinate is in the normal direction to the plane (positive northward), and the z coordinate is along the dip direction (positive upward). Figure 15c shows projections onto the x-y plane for z coordinate between 13.5 m to 34.5 m in steps of 3 m thickness in the z direction. Blue dots are clustered AE events (approximately 4900) defined at the plane, which are situated in the red dashed frames in (a) and (c). Gray dots are other events, which belongs not to the cluster. The AE hypocenters projected onto x–y planes exhibit clear traces continuous over several meters or more. It was shown that the events of this cluster correspond to the Zebra fault, a local fault.

**Figure 15.** Close up of area marked by Box B in Figure 13. The relocated AE events in projections onto the x-y plane (**a**) and x-z plane (**b**) are shown. A local coordinate system aligned with clear discernible plane is used. x coordinate is along strike direction (positive eastward), y coordinate is in normal direction to the plane (positive northward), and z coordinate is along dip direction (positive upward). (**c**) Shows projections onto the x–y plane for z coordinate between 13.5 m to 34.5 m in steps of 3 m thickness in the z direction. Blue dots are clustered, which are located in the red dashed frames in (**a**,**c**). Gray dots are other events [84].

The information gained from studying the characteristics of AE event on such structures revealed in-depth information useful for risk assessment and stope planning. For example, the formation of Ortlepp shear fractures [95,96] due to the approaching stope front could be observed (see Figure 14b); AE activity picking up on the newly created formation already, when the mining front was still more than 20 m away [34].

At the same time, pre-existing discontinuities such as faults or dikes that get seismically activated by the stress re-distribution due to the excavation were also identified [34,84]. Higher-order features such as branches and step-overs could be observed owing to the high localization certainty. Detailed analysis of AE event properties including a b-value study revealed that both quasi-static slip [85] as well as dynamic slip events is observed on faults [84].

Interestingly the frequent observation of repeating AE events on faults monitored during the SATRAPS project suggests that AE events on faults loaded by stress-redistribution due to mining undergo a similar process as tectonic faults subject to tectonic stresses [86]. Note that both Ortlepp shear fractures as well as the re-activation of faults cause on a regular basis violent rock burst events (M1 to M4), which are a significant threat to people working underground [96–98].

#### *4.2. Gold Mine Mponeng in South Africa*

The joint Japanese-German Underground Acoustic Emission Research project (JAGUARS) (see Table 3) in South Africa measures AE events in the frequency range from 700 Hz to 200 kHz. In the JAGUARS project [73] conducted in Mponeng Gold Mine in Carltonville in South Africa in 3.3 km depth the full evolution of a rock burst with moment magnitude *MW* = 1.9 could be monitored using an in situ AE monitoring network. The JAGUARS network was installed in spring 2007.


**Table 3.** Group members of the joint Japanese-German Underground Acoustic Emission Research in South Africa (JAGUARS) project [36,73].

The gold is mined from the Witwatersrand formation. The gold-carrying sedimentary layer (Ventersdorp Contact Reef) is embedded in a thick series of quartzite, dipping with 26.5◦ toward the south-east, and reaches a thickness of 0.5 to 1 m. The JAGUARS network is located approximately 90 m below the reef, next to a gabbroic dike (Pink-Green (PG) dike, Figure 16). This dominant geological feature is 30 m wide and dips nearly vertically. It cuts through the reef and serves as a support pillar for the exploitation.

Naoi et al. [99] estimated the seismic velocities of rock types in the vicinity of the network from velocity tomography using AE transmission measurements. For the quartzite host rock, Naoi et al. [99] estimated seismic velocities of *vP* = 6.2 km/s for P waves and *vS* = 3.8 km/s for S waves. The velocities within the pink-green dike were found to be slightly higher, with *vP* = 6.9 km/s and *vS* = 3.9 km/s.

Mining in the vicinity of the JAGUARS network (Figure 16) started in early 2007. The network focuses on the dike-host rock contact close to the mining front, where larger events with magnitudes up to *MW* = 3.0 were expected due to stress concentration induced by mining (personal conversation with S. Spottiswoode, 2009). The seismic network is consisting of 8 borehole AE sensors and one triaxial accelerometer. All sensors were installed in short boreholes of 6 m to 15 m length. The data were recorded in trigger-mode recording.

**Figure 16.** Sketch of the location of the JAGUARS network in the Mponeng gold mine and its surrounding geologic formations. The JAGUARS network is located at 3540 m depth. The sensors are located in boreholes, that are shown as bold black lines. The Pink-Green (PG) dike is shown in gray. The gold reef with active mining located above the JAGUARS network is shown in light gray. The mining stope width is 0.5 m to 1 m. The development tunnels XC45 and XC46 are shown in dark gray [52]. Reproduced with permission.

The stress changes owing to the approaching mining front on the PG dyke are modeled by Ziegler [77]. A main shock with magnitude of *MW* = 1.9 occurred on 27 December 2007 in the center of the AE monitoring network [33,36,74]. Naoi et al. [74] manually picked P- and S-wave arrival times to locate more than 20,000 AE events, that occurred within 150 hours, following the main shock. The location error for events within a radius of about 40 m of the center of the AE network was less than 1 m. Most of the AE events from this period occurred within 50 m to 100 m of the network, where the spatial coverage of the network is best. The events contain signals with a broad range of high frequencies, which allowed the sensitivity of the network toward very high frequencies above 25 kHz to be analysed in greater detail [36,52]. Owing to the resulting excellent recording of the aftershock AE event sequence with most AE events after the main shock, the rupture process and the rupture plane could be studied in great detail.

The spatial distribution of 9,444 aftershocks is shown in map view in Figure 17a. The main shock hypocenter is shown by a grey star. The positions of cross sections I to IV are shown by magenta rectangulars. In Figure 17b the cross sections are presented the location of manual re-located AE events in a side-view. The PG dyke is shown by two grey lines. The re-located events reveal that the rupture plane started within the PG dyke, but reached the geological boundary, where branching and bending occurred. Secondary features (marked with VII and IX) are observed that likely correspond to the rupture of major aftershocks.

**Figure 17.** Mainshock-Aftershock sequence from 27 December 2007 recorded by the JAGUARS network. The spatial distribution of 9,444 aftershocks is shown in map view in (**a**). These aftershocks comprise a subset of the 25,000 aftershocks recorded which come with small residuals of the automatic locations. The mainshock location is shown by a grey star. Red stars show the location of all aftershocks recorded by the in-mine geophone network. The event depth (shaft depth) is color-coded, which shows that aftershocks were triggered not only on rupture plane, but also at the stope face above the mainshock. The positions of cross sections I to IV are shown by magenta rectangulars. In (**b**) cross sections are presented that show the location of manual re-located AE events in side-view. The PG dyke is shown by two grey lines.

Yabe et al. [78] showed that the aftershock AE events clearly delineated a plane in the PG dike with a strike of N22W and a dip of 68◦ toward N68E (Figure 17). Because waveforms of the main shock recorded by the AE network were saturated in AE recordings, the main shock was analyzed using waveform data of the in-mine geophone network. Naoi et al. [74] were able to resolve the complexity of the rupture plane (approximately extension 100 m × 80 m), which underwent branching and bending according to geological heterogeneities present. Naoi et al. [74] applied the master-event location technique to locate the hypocenter of the main shock relative to the aftershock AE events. The main shock hypocenter obtained was about 30 m above the AE network and on the aftershock plane. The focal mechanism solution of the normal fault for the main shock using seismic waveforms recorded by the seismic network operated by the mine pointed out that one of the nodal planes agreed well with the aftershock plane. Therefore, the aftershock plane is considered to correspond to the rupture plane of the main shock plane, which demonstrates clearly that the AE aftershock activity outlines the main shock's rupture plane. Kozlowska [79] show that the aftershocks occur in areas of positive Coulomb stress change as determined using rate and state based stress modeling.

Source parameter analysis demonstrated that the aftershock AE event sequence has the same characteristics as tectonic aftershock events, i.e., they follow the Omori law [36], the Gutenberg–Richter distribution [55] and a static stress drop [16]. The magnitude ranges between −5.0 and −0.8. The magnitude of completeness varied strongly in space, but was estimated to *MC* = −4.8 in the network center [52].

The area of rupture initiation was subject to foreshock activity that is interpreted as the breakdown of asperities [78]. Interestingly, there are indications that AE events announce the main shock in advance. Figure 18a,b display the results of manual and automatic source location, respectively, applied to the same AE events, which are within 5 m of the aftershock surface. The events occurred during the periods before the main shock and within 150 h following the main shock. The black dashed contour represents the area of significant aftershock activity, as defined by Naoi et al. [74]. As this area can be presumed to represent the rupture area of the main shock, AE events that took place in the aftershock area and within 5 m of the aftershock surface before the main shock are hereafter referred to as "foreshocks". Four foreshock Clusters F1 to F4, and two aftershock Clusters A1 and A2 were identified from a concentration of manually located AE events. Both the foreshock and the aftershock clusters barely overlap one another. For comparison, the automatically located AE events, which are much more numerous than manually located ones are shown in Figure 18b as a density plot.

**Figure 18.** Foreshock activity compared to aftershock activity. (**a**) Distribution of the manually located AE events that occurred within 5 m of the aftershock surface during the period from 13 June 2007 to 150 h after the main shock. The yellow star indicates the main shock hypocenter. Black, red, blue, and green solid circles denote the events in June, September, October, and December, respectively. Orange and black thick solid lines enclose the foreshock clusters (F1, F2, and F3) and aftershock clusters (A1 and A2), respectively. Gray dots denote the aftershocks. Gray contours show the areal density of the aftershocks drawn by feeding areal densities in 5 m× 5m cells. The thin black dashed contour indicates the aftershock area defined by Naoi et al. [78]. The light blue solid circle is the access tunnel along which the observation network was deployed. (**b**) Distribution of the automatically located events that occurred before the main shock and within 5 m of the aftershock surface. Their densities in 5 m × 5 m cells are shown by gray scale. Blue, purple, and green solid circles denote events in October, November, and December, respectively. Light green contours show the areal density of the automatically located foreshocks.

#### **5. In Situ AE Monitoring During Hydraulic Fracturing in Mines**

A slightly different aspect of SHM concerns the monitoring of hydraulic fracturing (HF) using in situ AE monitoring. Engineered fractures generated underground by packer probes in boreholes are facilitated in a broad variety of contexts, many of which require detailed knowledge on the damage process actively initiated. HF is a common tool for underground stress determination and provides important input for designing the stope layout and for risk assessment [100–102]. HF has become a widely used engineering tool in reservoir enhancement of geothermal systems, shale gas, or conventional oil and gas extraction as it effectively increases the permeability [103–106]. In addition, HF is successful in increasing the productivity in ore production e.g., HF is used in fragmenting ore bodies [107].

Several research projects have addressed HF using in situ AE monitoring or strain cells in order to increase the understanding of the rock response to HF, to study the evolution of fracture generating and predict the stimulation of existing fractures [67,108,109]. Two recent research projects in crystalline rock did not only record very interesting and rich data, but have pushed the limits in highly sensitive monitoring and advanced signal processing. Zang et al. [43] report on HF at the hard rock underground laboratory Äspö in Sweden (Figure 19a).

**Figure 19.** (**a**) Test site of hydraulic fracturing tests in the underground Äspö Hard Rock Laboratory in Sweden (see http://www.skb.se/upload/publications/pdf/Aspo\_Laboratory.pdf). (**b**) Location of the AE borehole sensors together with the central injection well (blue line). The blue star identifies the fluid injection segment corresponding to the hydraulic fracturing (HF2) experiment [43,87].

Zang et al. [43] utilize three different monitoring networks, namely in situ AE monitoring, microseismic monitoring, and electromagnetic monitoring. Figure 19b shows a top view of the AE monitoring network, which consists of 11 AE sensors of GMuG type MA BLw-7-70-75 and four Wilcoxon 736T accelerometers.

Data is recorded both in trigger mode and continuous mode (1 MHz sampling frequencies). AE sensors are installed in monitoring boreholes of 22 m to 30 m length parallel to the injection borehole and in short boreholes along the tunnels.

Overall, six HF stimulations are performed using three different injection schemes (continuous, progressive and pulse pressurization), from which four produced significant AE activity outlining the fracture orientation, extension and temporal evolution. Within 20 days, about 69,400 triggers were recorded in situ, from which many correspond to noise events due to dripping water or anthropogenic activities. The strongest AE events recorded in situ with best signal-to-noise ratio and most reliable location certainty (maximum location residual 0.3 m) formed a catalog of 196 seismic events, of which all correlated in time and space directly to the HF periods. All relocated AE events are shown in a perspective view (Figure 20a) and a lateral view in a rotated coordinate system (Figure 20b) for seismically active HF stimulations (HF1 to HF4 and HF6). AE events recorded during the different fracture experiments clearly delineate the fractures and display differences between the different hydraulic fractures generated.

A faster and further expansion of AE events away from the stimulation point is observed with each subsequent re-stimulation stage outlining the damage extension. In-depth source analysis of the largest AE events including energy estimation, moment tensor inversion, source parameter estimation, and stress inversion by Kwiatek et al. [27] estimated the moment magnitude *MW* of the AE events ranges from −4.2 to −3.5. The source analysis clearly reveals that most events correspond to shear slip events on pre-existing fractures. AE event activity starts as soon as a certain pressure level is reached. First optimum oriented fault planes fail but, overall, differently oriented fault planes are observed. Stress inversion reveals stress rotation during and after the stimulation.

**Figure 20.** Location of AE events of six HF stimulations (HF1 to HF4 and HF6) presented in a top view (**a**) and a lateral view (**b**). For lateral view the coordinate system was rotated in that way to indicate the preferred fracture traces. The solid grey line outlines injection well and the observation boreholes (modified from [43]). All dimensions are given in meter.

Using the continuous data recordings from the Äspö experiment, López-Comino et al. [87] demonstrated that using automated full waveform detection algorithm during post-processing could significantly increase the amount of triggered AE events. No seismic fracture event was recorded by the microseismic monitoring network or the Wilcoxon accelerometers, although the latter is capable of measuring the frequency range of the observed AE events. Owing to the small nature of all AE events recorded, not only seismic monitoring in the kHz range was essential, but also the significantly higher sensitivity of the in situ AE sensors.

A different experiment was conducted at the Grimsel Test Site (GTS) in Switzerland (Figure 21a,b) operated by the Swiss National Cooperative for the Disposal of Radioactive Waste (Nagra). The GTS is located at 1,733 m above the sea level and has an overburden of 400 to 500 m. Gischig et al. [54] implement in situ AE monitoring for stress determination by HF in the Grimsel site. The so-called in situ stimulation and circulation (ISC) [110] was performed between two tunnels i.e., the VE and the AU tunnel (see Figure 21c), and the injection and monitoring boreholes were mostly drilled from the AU cavern at the southern end of the AU tunnel (Figure 21c). The host rock is the so-called Grimsel granodiorite, which changes into the Central Aar granite about 50 m north of the experiment volume [111]. The rock mass in the experiment volume is exceptionally intact. The Grimsel test site was monitored using 28 AE sensors (type GMuG MA-Bls-7-70) and four Wilcoxon accelerometers. Most AE sensors were installed on the tunnel wall on polished rock face, while eight AE sensors were installed in a water-filled vertical borehole (Borehole SBH1 in Figure 21c).

For stress determination, a series of hydraulic fracturing tests and overcoring were performed. During hydraulic fracturing, nearly 2000 AE events were recorded with a source-receiver distances smaller than 30 m that outlined, similar to the Äspö experiment, clearly the fracture plane that extended

up to 5 m from the injection point (see Figure 22). Events occurred mostly during the refracturing cycles once a critical injection volume of 0.5 to 1 liter was exceeded and less during the initial fracturing cycle. A comparison of the fracture plane outlined by AE events and stress measurements using an imprint packer and overcoring testing, revealed significant deviations. The imprint packer revealed that fractures initiated at the borehole wall within the foliation plane, but the fracture growth than rotated, as outlined by AE events, in such way that it extends normal to the minimum principal stress. The deviation of the overcoring stress measurement result to the actual fracture plane observed could be explained by using a transversely isotropic elasticity model.

**Figure 21.** (**a**,**b**) Grimsel test site is located in the Bernese Alps in southern Switzerland (see www.grimselstrom.ch). (**c**) In situ AE monitoring was performed during hydraulic fracturing tests HF1 to HF3 in Borehole SBH3. Positions S1 to S28 mark the location of the AE sensors and accelerometers [88].

The authors conclude that AE monitoring was crucial for the combined interpretation of the stress characterization results and to maintain meaningful stress estimation.

We summarize that HF routine monitoring using in situ AE monitoring systems becomes feasible for underground production, if sensitive AE sensors for in situ operation are used. Two upcoming projects performing underground medium-scale HF testing will implement AE monitoring accordingly: the enhanced geothermal system (EGS) Collab project's stimulation experiment in Sanford Underground Research Facility in the former Homestake Gold mine, USA [112] and the STIMTEC stimulation experiment in the underground laboratory in the silver mine "Reiche Zeche" in Germany (personal conversation with J. Renner and G. Dresen, 2018) (more details at http://stimtec.rub.de).

**Figure 22.** Perspective view of AE events detected during hydraulic fracturing tests HF1 to HF3. The continuous line indicates the injection well with the position of the injection intervals [54].

#### **6. Concluding Remarks**

In conclusion, this article summarizes the capability of in situ AE monitoring in the context of SHM based on the results of monitoring projects in mines. The in situ AE method is capable of detecting microcracking, in high resolution and sensitivity, which is caused by very small deformation processes at high deviatoric stresses. This means that in situ AE monitoring provides detailed insights into the ongoing deformation processes.

In contrast to in situ AE monitoring, microseismic monitoring is used to measure large-scale deformations in mines, which may cause rock bursts or roof falls. Due to limitations in frequency range and sensitivity, microseismic networks are not able to detect microcracks. Therefore, small AE events are very often not considered for stability assessment and interpretation of geomechanical conditions of the rock. This work clearly shows that in situ AE monitoring is able to detect very small AE events in zones of weakness related to dynamic processes like dilatation. Therefore, in situ AE monitoring is a useful tool to monitor the geomechanical conditions of the host rock.

Real-time processing gives direct information on the location of AE events as well as on clustering, migration of AE activity or aftershock sequences of microseismic events. Recent advances in computer storage capacity allow recording of continuous data streams with 1 MHz sampling in addition to trigger mode recording, which makes advanced post-processing techniques possible.

The results shown here also demonstrate that monitoring of larger rock volumes with in situ AE measurements is possible in various rock types. Detections of AE events from distances much greater than 100 m is possible in rock with low wave attenuation like salt rock or hard rock. In this case, rock volumes far away from the AE network can be monitored. On the other hand, the in situ AE monitoring method is able to identify "aseismic" zones because AE activity is expected during significant damage processes. Finally, in situ AE monitoring is capable of detecting zones in mines where instability may appears long before macroscopic damage becomes visible, which is the objective of SHM.

**Author Contributions:** G.M. and K.P. wrote the article.

**Acknowledgments:** The authors thank K. Ono and two anonymous reviewers for their comments and suggestions to improve this article. The authors are grateful to T. Spies and D. Kaiser (BGR in Hannover), G. Kwiatek (GFZ in Potsdam), and H. Moriya (Tohuko University in Sendai) for providing updated figures for this review.

**Conflicts of Interest:** The authors declare no conflicts of interest.
