Available online at www.sciencedirect.

com

ScienceDirect
Procedia Engineering 191 (2017) 697 – 702

Symposium of the International Society for Rock Mechanics

Automated Full Waveform Detection and Location Algorithm of

Abstract

A near field network with 11 acoustic emission (AE) sensors was installed for the in situ underground experiment (Nova project
54-14-1) that took place 410 m below surface in the Äspö Hard Rock Laboratory, Sweden. The acquisition system for the piezo-
electrical sensors has been improved to record signals with 1 MHz sampling rate, to detect signals produced by weaker sources
and enhance the microseismic catalogue. The acquisition system was capable to operate in trigger and continuous mode.
The basic idea of the experiment was to compare hydraulic fracturing growth and induced seismicity under controlled conditions
for different loading scenarios as conventional versus progressive, and pulse-like water injections. In this work, we consider
continuous recordings and apply recently developed automated full waveform detection and location algorithms which are based
on the stacking of characteristic functions calculated from squared amplitudes. Waveform stacking and coherence techniques are
adapted to detect and locate AE signals for massive datasets with extremely high sampling. We significantly increase
the detection rate in comparison to trigger mode routines. Most detection concentrated during the fluid injection occurred around
the fracking stages. Frequency-magnitude distribution characteristics are investigated using a relative magnitude scale estimated
from the amplitude recorded at AE sensors. We demonstrate that the stacking of characteristic functions yields to a significant
improvement of the detection and location also in presence of noisy records, supporting the adoption of similar techniques for
other induced and natural seismic activity monitoring systems.
© 2017
© 2017TheTheAuthors.
Authors. Published
Published by by Elsevier
Elsevier Ltd. Ltd.
This is an open access article under the CC BY-NC-ND license
Peer-review under responsibility of the organizing committee of EUROCK 2017.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of EUROCK 2017
Keywords: Hydraulic fracturing, induced seismicity, acoustic emission, full waveform detection, detection performance

* Corresponding author. Tel.: +49-331-288-1325; fax: +49-331-288-1127.

1877-7058 © 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of EUROCK 2017
doi:10.1016/j.proeng.2017.05.234
698 José Ángel López Comino et al. / Procedia Engineering 191 (2017) 697 – 702

1. Introduction

An in situ hydraulic fracturing experiment was performed at Äspö Hard Rock Laboratory (Sweden) aiming at
optimizing geothermal heat exchange in crystalline rock mass [1]. A near field network with 11 acoustic emission
(AE) sensors was installed 410 m below surface to map the seismic response of hydraulic fractures for different fluid
injection scenarios (Figure 1). The location of the experimental tunnel TASN is seen from which four long boreholes
were drilled sub-parallel to orientation of minimum horizontal compressive stress. The middle borehole (Figure 1,
blue line) serves as hydraulic testing borehole and was drilled to a total length of 28.40 meter, down dipping -4°.
The rest of monitoring boreholes were drilled with inclination upwards to allow water outflow from AE sensor
chains. This geometry in the predetermined stress state allows propagating hydraulic fractures perpendicular to
the hydraulic testing borehole, in the direction of maximum horizontal compressive stress, and in a direction towards
the monitoring boreholes.

Fig. 1. Test site for hydraulic fracturing in an experimental tunnel of Äspö Hard Rock Laboratory, Sweden (left). [elaborated after
http://www.skb.se/upload/publications/pdf/Aspo_Laboratory.pdf ]. Sensors are employed in the near-field (right): a blue line indicates
the hydraulic testing borehole, the blue star identifies the fluid injection segment corresponding to the HF2 experiment.

The basic idea of the experiment was to compare hydraulic fracturing growth and induced seismicity under
controlled conditions in a horizontal borehole 30 meter long for continuous fluid injection versus progressive fluid
injection, and dynamic pulse hydraulic fracturing [1]. The in-situ AE monitoring network consists of eleven AE
sensors (GMuG MA BLw-7-70-75) and four accelerometers (Wilcoxon 736T). AE sensors employed are uniaxial
side-view sensors for borehole installation, developed by GMuG for sensitive recording in the frequency range
1 kHz to 100 kHz and capable to monitor fractures from centimeter to meter scale. Sampling rates were extended to
1 MHz and all sensors are installed inside boreholes. Eight AE sensors are installed in long monitoring boreholes,
surrounding the fracturing borehole. The remaining sensors are installed in short boreholes near the tunnel roof. Data
was recorded using the measuring system GMuG AE-System that is suitable both for continuous recording of data
and recording in trigger mode. In this work, we present the results obtained during the conventional, continuous
water-injection experiment HF2 (Hydraulic Fracture 2) in Ävrö granodiorite, and discuss the detection performance
using a recently developed automated full waveform detection algorithm.
 José Ángel López Comino et al. / Procedia Engineering 191 (2017) 697 – 702 699

2. Automated full waveform detection

We consider continuous recordings and apply a recently developed automated full waveform detection and
location algorithms (Lassie [2]). Lassie has been extended for this specific dataset to analyze extremely high
sampled data (1 MHz). Lassie is a python-based earthquake detector, which relies on the stacking of characteristic
functions. It follows a delay-and-stack approach, where the likelihood of the hypocentral location in a chosen
seismogenic volume is mapped by assessing the coherence of arrival times at different stations (see [3] for
an overview). However, in the Lassie implementation, the adoption of smooth characteristic function calculated
from normalized amplitude envelopes allows to reduce the spatial and temporal sampling. This improves
the computational performance of the algorithm and allows its application to high-sampling data as a detector.
The outstanding computational performance and smooth imaging of the coherence function are achieved at the cost
of a larger location uncertainties which accuracy can still be improved upon each event detection by applying
different characteristic functions [4, 5].
Preliminary locations are found for each detected event according the maximum value of coherence and
the travel-time stacking corrected with P and S-wave velocities. We have considered a homogeneous full space
model where the velocities for both P- and S-waves were obtained from active ultrasonic transmission tests
(vp= 5810 ± 120 m/s and vs = 3400 ± 200 m/s). The waveform signature of most noise events differ clearly from
seismic events, however, a visual inspection of seismic waveforms reveals different kinds of detected signals that we
have classified as: acoustic emission (AE) detections (Figure 2), false detections, electronic noise, anthropogenic
noise, long period noise and other signals (Table 1 and Figure 3).

Fig. 2. Typical waveform of seismic Acoustic Emission (AE) events detected by the Lassie algorithm [2] during HF2, showing timing and
amplitude of the characteristic function (Acf). Waveforms are band-pass filtered in the frequency range 3 – 70 kHz. The time (ms) is shown on
the x-axis and the reference time is displayed in the lower left corner of each box.

Table 1. Class and number of detected events in the Hydraulic Fracture 2 (HF2) experiment.
Class of detected events Number of detections
Acoustic emissions detections 4179
False detections 4158
Electronic noise 18894
Long period noise 361
Anthropogenic noise 58
Other signals 215
700 José Ángel López Comino et al. / Procedia Engineering 191 (2017) 697 – 702

3. Detection performance

The detection rate results highly discontinuous, with the highest AE detection rates well corresponding to all fluid
injection stages (Figure 3). Three AE events are detected during the packer inflation before the initiation of HF2.
A rapid microseismicity increase is experimented when the constant injection rate is applied, afterwards, a rapid
decrease is observed corresponding with few and isolated events. The number of AE events is accounted for
the different stages of HF2 (table 2), being the Refrac 1 and 5 where the microseismicity is larger. Note that
the continuous AE recording was not in operation during the whole experiment Refrac 2 and finished prematurely.
This explains the detection of only 4 events during this stage.

Table 2. Number of acoustic emissions events during the different stages for the Hydraulic Fracture 2 (HF2) experiment.
Experiment - HF2 Number of AE events
Pulse Test 3
Frac 690
Refrac 1 1325
Refrac 2 4
Refrac 3 551
Refrac 4 595
Refrac 5 990

A low detector threshold is chosen, in order not to loose weaker events and reach a complete catalogue. However,
the small threshold also increases the number of false detections requiring an accurate classification. The dataset has
been revised manually, and detected events classified in terms of real AE events, electronic noise, anthropogenic
noise, long period noise and other signals. The waveform signatures of most noise events differ clearly from seismic
events and were identified using recording in trigger mode for (near) real time assessment (see noise examples in
[6]). Electronic noise is found temporally associated with the fluid injection stages. Its occurrence hinders the search
of real events. These signals share the same arrivals in all traces and similar frequency content and duration.
Classification algorithms matching waveform or spectral patterns may be used for future analysis, but are not
considered here. However, noise events generated by people working near the network are difficult to identify
because they display a transient character like seismic events, especially when they were generated by (hammer)
blows to the rock wall or dropping tools. A careful visual inspection is required to consider these aspects. Long
period noise corresponds with harmonic signals originates inside the monitoring borehole and is presumably due to
the irregular water inflow. We found other signals that can not fit in the previous families and do not share similar
characteristics to establish other possible families. Most of them are detected after the well is shut-in and could also
be caused by human factors. However, some of them are observed during the constant injection rate is applied, in
that case, they may indicate overlap of electronic noise and possible real events, which hinders a safe classification.
Excluding the AE signals and electronic noise, for the remaining types of signals the amplitude of the characteristic
function are usually low, so that these events can be easily removed by increasing the detection threshold (e.g. to
an amplitude of characteristic function value of 750, as shown in Figure 3).
 José Ángel López Comino et al. / Procedia Engineering 191 (2017) 697 – 702 701

Fig. 3. Acoustic emission detections (red dots) using continuous recordings and other kinds of detected signals from a visual inspection that do
not correspond with seismic events: false detections (gray dots), electronic noise (orange dots), anthropogenic noise (green dots), long period
noise (black dots) and other signals (magenta dots). Amplitude of the characteristic function (left ordinate) calculated from Lassie detector,
injection pressure (black line) and flow rate (blue line) for hydraulic fractures (right ordinate) in the experiment HF2 are shown. Different stages
for HF2 are indicated using gray background (PT: Pulse test; F: Frac; RF1 - RF5: Refrac 1 - Refrac 5).

4. Conclusions

Based on the analysis of acoustic emission events from continuous recording in an underground experiment with
hydraulic fracture growth in naturally fractured crystalline rock, we draw the following conclusions:

x We adapted and successfully used for the first time a full-waveform based detector, which relies on
the location-based stacking of smooth characteristic functions to detect AE signals for massive datasets with
large number of sensors and/or extremely high sampling (here 1 MHz). Using continuous waveforms we are
able to identify a large number of events, even in presence of microseismicity bursts, when multiple events
occur close in space and time. Moreover, our approach is based on the detection of coherent increased in
waveform amplitude at multiple sensors, being able to detect events even in presence of noisy data. At this
respect it should be noted that a triggered based approach [1] was able to detect 102 events in the same
dataset, whereas our catalogue is finally composed of 4158 events. The better detection performance leads to
a relevant increment of the catalogue size, implying a decrease of the magnitude of completeness, Our results
extend the adoption of similar detection techniques, successfully applied for monitoring induced and natural
seismic activity at local and regional distances [7,8], also to small-scale applications and for hydraulic
fracturing.
x The detection threshold is the most relevant parameter which influences the detection performance of the used
algorithm. A low threshold allows the detection of weak events at the cost of a higher number of false
detection and noise signals of different types. In this sense, it is desirable that the detection setup can be
combined with a classification algorithm (here we showed results of a visual classification) to distinguish true
and false events, and classify different signals in an automatic manner. The inspection of the temporal
evolution of signal detection reveals that 85% of AEs take place during the phases of increased flow rate and
increasing pressure, dropping very quickly in time as soon as the pressure decrease and the flow stopped. In
the time periods between each fracture and refrature operations, very few and weak events are detected.
702 José Ángel López Comino et al. / Procedia Engineering 191 (2017) 697 – 702

Acknowledgements

This work is funded by the EU H2020 SHEER project (www.sheerproject.eu). Nova project 54-14-1 was
financially supported by the GFZ German Research Center for Geosciences (75%), the KIT Karlsruhe Institute of
Technology (15%) and the Nova Center for University Studies, Research and Development (10%). An additional in-
kind contribution of SKB for using Äspö Hard Rock Laboratory as test site for geothermal research is greatly
acknowledged.

References

[1] A. Zang, O. Stephansson, L. Stenberg, K. Plenkers, C. Milkereit, G. Kwiatek, G. Dresen, E. Schill, G. Zimmermann, T. Dahm, M. Weber,
Hydraulic fracture monitoring in crystalline rock mass at 410 m depth using an advanced fluid-injection protocol and extensive sensor array,
Geophys. J. Int. 208(2) (2017) 790–813.
[2] S. Heimann, C. Matos, S. Cesca, I. Rio, S. Custodio, Lassie: A versatile tool to detect and locate seismic activity, Seismological Research
Letters, in preparation. Note: interested users to preview Lassie can write to: [email protected], 2016.
[3] Cesca, S. and F. Grigoli (2015). Chapter Two – Full Waveform Seismological Advances for Microseismic Monitoring. - Advances in
Geophysics, 56, p. 169–228.
[4] F. Grigoli, S. Cesca, M. Vassallo, T. Dahm, Automated seismic event location by traveltime stacking: An application to mining induced
seismicity, Seismol. Res. Lett. 84 (4) (2013) 666–677.
[5] F. Grigoli, S. Cesca, O. Amoroso, A. Emolo, A. Zollo, T. Dahm, Automated seismic event location by waveform coherence analysis,
Geophys. J. Int. 196 (3) (2014) 1742–1753.
[6] Plenkers, K. and T. Fischer, Field report: GMuG Acoustic emissions monitoring in FHF experiment, Äspö underground laboratory,
Oscarshamn, Sweden, 29 May to 22 June, 2015.
[7] Matos, C., S.,Heimann, F. Grigoli, S. Cesca and S. Custódio, Seismicity of a slow deforming environment: Alentejo, south Portugal, EGU
General Assembly 2016, EGU2016-278.
[8] López Comino, J. A., S. Cesca, M. Kriegerowski, S. Heimann, T. Dahm, J. Mirek and S. Lasocki, Monitoring performance using synthetic
data for induced microseismicity by hydrofracking at the Wysin site (Poland), Geophys. J. Int. (2017), under review.