Get Complete Project Material File(s) Now! »
Seismic activity and observed damage
Up to now, no rockbursts neither other major failures have been observed in the studied area of Lappberget orebody. Most of visible damage occur within stope openings or in galleries at the opening back (where the opening back refers to the gallery of an open stope). Other types of damage can be stress-related or linked to weakness zones.
Boliden has produced an internal database of the main microseismic events recorded in Garpenberg mine from 2009 until 2016, associating the events to the damage observed on site. Seismic data were recorded by a monitoring network installed by the Institute of Mine Seismology (IMS) in a shallower zone of Lappberget. If interested in the description of this network, the reader is referred to Olivier et al. (2015). According to the database, Lappberget experienced 59 significant seismic events from August 2012 until January 2016, which were located in the upper part of the orebody, between levels 464 and 993. Even if these events are outside the study area of this thesis, their analysis can give valuable information about the rock mass response to mining in Lappberget orebody.
Among the total dataset, we observe 52 major events with local magnitude (ML) ranging between 0 and 1.5, while the remaining are smaller events (-1.8 < ML < 0) which lead to damage. As reported in Fig. 2.15, most of major events have ML ≤ 0.6 and they are mostly located within 10 meters from the excavated front. Smaller events have similar distances from excavations, except one event which is located 40 m away.
As suggest by the substantial number of events at a distance from the excavation equal to 0 (Fig. 2.15b), damage mostly occurred within stope openings. An example of this type of failures is given by the event that occurred on 2015-07-22 in stope 5 at level 602. The event, with ML = 0.3, was felt at the surface and it resulted in the widening of the excavated stope (Fig. 2.16a). This failure was most probably due to high stresses and because of the presence of a weak zone near the excavated area (Mozaffari 2016, personal communication, 15 February).
Beside stope openings, most of the failures can be observed within drifts at the opening back or in adjacent galleries. This mainly results in the occurrence of fractures and/or falls of rock and shotcrete (Fig. 2.16b), broken bolts and water leakage. In the end of 2011 a strain burst occurred at level 896 of Lappberget, causing important damage as, for example, rock falls, cracks on the floor and heaving of new and existent fractures. Boliden classified the damage as level 4 on 5 for, both, rock and support structures, based on the rock and support damage scales proposed by Kaiser et al. (1992). Four years after the event, this area of the mine presents deterioration of pillars in the footwall drift, floor heave, and water leakage, which indicates the presence of fractures (Fig. 2.16c and d). No other seismic events related to this area were reported in the Boliden database, but it is obvious that stress related deformations are occurring in the zone. Such deformations can be observed in many levels of Lappberget, in particular on footwall pillars. The explanation of these phenomena is that stresses are gradually distributed through the pillars due to adjacent stopes extraction, inducing irreversible strains, or plasticity, when the elastic capacity of the rock is exceeded (van Koppen, 2008).
Routines of seismic data acquisition and processing
As already introduced in Chapter 2, a permanent microseismic monitoring network was installed in block 1250 of Lappberget orebody, between December 2014 and February 2015. The network consists of 6 1C and 5 3C 14 Hz geophones, installed in 5-to-20-meter-deep, vertically-oriented or inclined boreholes, covering a monitored area with a volume of about 64×106 m3 between -950 m and -1350 m of depth (Fig. 3.1). Geophones are deployed over three mine levels (1108, 1182 and 1257) referred as A, B and C, where also acquisition units (AU) for data transfer are installed (see Tab. A.2 for exact probe positions and orientations). Timestamping is ensured with a GPS installed at the surface and linked by Ethernet connection to each AU, with a resolution of 1 ms.
The acquisition system is operating in triggering mode with 8 kHz sampling frequency, while the triggering threshold, based on amplitudes, is fixed at 200 digits (≈ 5.35×10-6 m/s). Moreover, it has been imposed that an event should be retained only if it triggers at least 2 probes per each AU on at least two levels. This results in increasing at 4 the minimum number of probes for a seismic signal to be retained. If less probes are triggered, the signal is considered as isolated and discarded for further analysis. This choice is motivated by monitoring needs. Indeed, as Lappberget is a production area, which is active 7 days a week and 365 days per year, many seismic signals are due to mine works (e.g. drilling, bolting, mucking, trucks or other vehicles). Therefore, limiting the minimum number of triggering probes should reduce the number of working-noise signals. However, this additional condition on the acquisition system has straightforward limitations for research needs, reducing significantly the number of detectable microseismic events (MSE), especially the smallest ones. Finally, a 1 kHz analog low-pass Butterworth filter is applied to the data at the moment of their acquisition.
Once detected, seismic signals are sent to the Ineris monitoring center where they are processed in semi real-time with the Ineris software suite SYTMIS, and searchable online via the interface e.cenaris. MSE localization is based on P an S-waves arrival times as well as on P polarization angles (dip and azimuth), which improves events localization when few seismic probes are available (Contrucci et al., 2010). However, polarization angles are only considered for 3C probes. A probabilistic approach is used to solve the inverse problem (Tarantola and Valette, 1982) and determine the hypocenter with the maximum likelihood. This consists in maximizing the probability density function (pdf) of the hypocenter by minimizing the misfit between observed and theoretical parameters, i.e. between observed and theoretical arrival times and polarization angles, using the L2- norm. The hypocenter is determined by the Oct-Tree nonlinear method (Lomax et al., 2000), which is a grid search technique based on a successive division of the space into cubes, depending on the pdf value per each cube. To apply this method, a homogeneous velocity model is assumed with VP = 6535 m/s and VS = 3703 m/s for P and S-waves respectively (VP/VS = 1.765), which was determined by calibration blasts and which agrees with the local geological context (Tonnellier et al., 2016). The searching space of the Oct-Tree algorithm is limited to a volume of 2000 m side along X, Y and Z directions, while successive cube dimension is ranging from 7 to 50 m. For taking into account velocity model uncertainties, an error of 0.002 m/s is fixed on travel times, while arrival times picking errors are set to 0.002 m/s and 0.003 m/s for P and S-wave respectively, which correspond to a location error of around 13 m considering P-wave velocity. Finally, polarization angles errors are set at 10° for azimuth and 15° for dip.
P and S-wave arrival times are determined by manual picking. In the case of 3C probes the S-wave picking is made simpler by means of waveform rotation (Abdul-Wahed et al., 2001; Cichowicz, 1993). Indeed, knowing probes orientation and assuming a straight ray path between source and receiver, the waveform can be rotated following the polarization direction of the signal. This means that the seismogram will be reoriented along three axes (LQT) which correspond to the P polarization direction (L-axis) and the two polarization directions of S-wave (Q and T axes). As a result, S-wave picking is facilitated on Q and/or T axes where P phase should appear significantly reduced (Fig. A.1). At the same time, P polarization angles are retrieved.
Besides events localizations, local magnitude (ML) and radiated seismic energy (E) are routinely determined as a standard procedure of daily data processing. Seismic source energy is estimated with a time-domain methodology, as the integral of the squared velocity seismogram and considering both P and S-waves. Only 3C probes are used for E estimation, whose final value is given by the arithmetic mean on all considered probes: 𝐸=4𝜋𝜌𝑁Σ𝑅𝑛2(𝑉𝑃∫𝑣𝑛(𝑡)2𝑑𝑡𝑡𝑆𝑛𝑡𝑃𝑛+𝑉𝑆∫𝑣𝑛(𝑡)2𝑑𝑡𝑡𝐸𝑛𝑡𝑆𝑛)𝑁𝑛=1(3.1).
where ρ = 3000 kg/m3 is the rock mass density determined by laboratory measurements on dry samples (Tonnellier et al., 2016), N the total number of 3C probes, Rn the hypocentral distance at probe n, VP and VS the P and S-wave velocity, tPn and tSn the P and S-wave arrival time at probe n, tEn the end time of the signal (95% of the emitted ES reached) and vn the ground velocity detected by the geophone.
Local magnitude of MSE is based on the computed radiated seismic energy. It was calibrated in order to be consistent with the ML estimated by another seismic monitoring network installed by the Institute of Mine Seismology (IMS) in the upper levels of Lappberget orebody, outside of the Ineris monitored area. This latter magnitude is estimated based on an empirical relationship derived from Vaal River and West Wits mines in South Africa, which takes the following form for Garpenberg mine: log10𝑃𝐺𝑉=0.58𝑀𝐿−1.85log10𝑟+1.05(3.2).
Type of recorded seismic signals
MSE recorded in Lappberget orebody between 2015 and 2016 present generally emergent onsets and a predominance of the S-phase (Fig. 3.2a, c, e, g). Signals are characterized by short durations, ranging from about 0.03 s to 0.7 s, and low amplitudes, generally smaller than 0.1 mm/s. For events characterized by high Signal-to-Noise-Ratios (SNR), P and S phases are generally distinguishable in the seismograms, even if the beginning of the S phase can be covered by the coda of P-wave, due to short travel distances. Three main classes of MSE can be distinguished according to their frequency content: (i) low frequency events with frequencies smaller than 500 Hz (Fig. 3.2a, b), (ii) large frequency band events, whose frequency intervals vary between 20 and more than 1000 Hz (Fig. 3.2c, d, e, f) and (iii) high frequency events with most of the energy distributed in frequencies which may be higher than 3000 Hz (Fig. 3.2g, h). A well-defined dominant frequency is not observed for any of the mentioned MSE classes, with intensities almost homogeneously distributed along frequency intervals. Due to the wide range of frequency bands, events reprocessed for the scope of this theses were corrected to remove the effect of the 1 kHz low-pass filter automatically applied to the data. The correction, required for source parameter estimation (Section 3.5), was applied within the band 200 ÷ 3000 Hz to avoid excessive distortion of original signals.
MSE characterization based on their frequency content highlights that rock mass extraction in Lappberget orebody may induce different event characterized by distinct dynamics. This first insight is interesting and will be later discussed in this thesis with the analysis of seismic source parameters which will give valuable information about the rock mass behavior in response to mining.
Besides MSE, a large number of signals recorded by the monitoring network are related to mine blasts and other types of mining operations. Indeed, as written at the beginning of this chapter, Lappberget orebody is currently under production, thus, it is characterized by intense blasting and working activities. Blast signals, related to PB and DB, are characterized by long duration, lasting in some cases more than 4 seconds, and by variable amplitude and frequency contents, depending on their distance from the seismic monitoring system. Indeed, blasts performed outside of the Ineris monitored area present lower frequencies (Fig. 3.3a, b), ranging between around 20 and 1000 Hz, than blasts performed within the monitored volume (Fig. 3.3c, d), which can show frequencies higher than 1000 Hz. Seismograms of these latter blasts are frequently saturated, showing a swarm-like shape due to a micro-delay between blasting of each explosive charge within separated drill holes (Fig. 3.3c).
Challenges and common errors in daily data processing
As highlighted in the previous section, signals recorded by the network are complex and related to various sources. An important task of the daily data treatment is the ability to recognize MSE and distinguish them from mine blasts and noises due to mine works, by visual inspection of seismic records. PB and DB are normally easy to identify, due to their unambiguous seismogram (Fig. 3.3a, c) and thanks to the regularity in blasting procedure (occurring at around 04:00 and 16:00 in local time). On the contrary, BB are much more difficult to distinguish from MSE as they can be performed at any time and only in few cases it is possible to be informed about their occurrence. In addition, their signals (Fig. 3.3e), in terms of waveform and duration, can be confused with that of MSE. Another difficulty comes from some signals related to mine works. Indeed, when these are not regularly repeated in time, as in the example of Fig. 3.3g, their waveform can look similar to that of MSE.
Beside waveforms classification, another challenge in the daily data treatment is due to picking of phase arrivals on triggered signals. Indeed, manual picking is often hard to perform, especially for MSE that present a low SNR. Moreover, when events are close to seismic probes, it can be difficult to distinguish between P and S arrivals, with S waves often covered by the coda of P ones. Time series picking can then be affected by different errors in arrival times identification and these errors are reflected on events location accuracy. These errors can be even more important if we consider that manual picking is performed by different operators.
With the aim of understanding the rock mass response to mining and identifying seismic hazardous areas for large damage prevention, a correct identification of real MSE related to induced rock mass fracturing and their accurate localization are two aspects that significantly influence the analysis of mining-induced seismicity. Indeed, the presence of blast and/or noise-related signals within MSE catalogues leads to erroneous interpretation of rock mass processes. At the same time, the wrong classification of MSE within noise and/or blast categories results in a loss of information which can be sometimes important. Once catalogues have been filtered from outliers due to “non-event” signals, the accurate localization of the real MSE is another basic concern. As reported by Mendecki et al. (1999), precise events localization is not only necessary for identifying rockbursts potential areas, but also because all subsequent interpretation and seismological processing, such as for example source parameters estimation, depend on location.
Based on these considerations and taking into account the complexity of Lappberget signals and the difficulties in manual picking, a back-analysis of seismic signals recorded from February 2015 to June 2016 was carried out. The aim of this work was, on one hand, the evaluation and correction of P and S phase picking and, on the other hand, a better signal classification to remove from the database all the recordings erroneously classified as MSE. Methodology and results of this analysis are discussed in the next section.
Picking consistency evaluation – The Wadati analysis
A suitable methodology for picking consistency evaluation is the Wadati diagram analysis (Wadati, 1933) which was already applied to earthquake data (Romano et al., 2013), as well as to mining-induced seismicity (Julià et al., 2009). Wadati diagram was originally introduced to estimate the origin time (t0) of seismic events, plotting on a diagram the phase arrival times difference (TS-TP) as a function of P arrival times (TP) for a single event recorded at different stations. To avoid working on single events, the modified Wadati diagram (Chatelain, 1978) was preferred as it allows grouping all the events in a single diagram, comparing time difference of P and S phases on station pairs.
Considering xi and xj as the hypocentral distances of a seismic event at two stations i and j, one can write: 𝛥𝑇𝑃=𝑡𝑃𝑖−𝑡𝑃𝑗=(𝑥𝑖−𝑥𝑗)𝑉𝑃⁄(3.4) 𝛥𝑇𝑆=𝑡𝑆𝑖−𝑡𝑆𝑗=(𝑥𝑖−𝑥𝑗)𝑉𝑆⁄(3.5).
where tP and tS are P and S waves’ arrival times, respectively, while VP and VS are waves’ velocities. Combining equations (3.4) and (3.5) we obtain: 𝛥𝑇𝑆=(𝑉𝑃𝑉𝑆)𝛥𝑇𝑃(3.6).
which represents the equation of a line with slope equal to VP/VS. Equation (3.6) is valid only assuming a propagating medium of constant velocity and, thus, a straight ray-paths joining sources and receivers. This assumption, even not true in a real geologic medium, is often employed in seismic analysis, introducing inevitable uncertainties which are although reduced when short source-receiver distances are considered (Julià et al., 2009), as in the case of Lappberget monitored area.
Table of contents :
Chapter 1 – Introduction
1.1 Mining-induced seismicity mechanisms
1.2 Monitoring of mining-induced seismicity and hazard assessment
1.3 Motivation, strategy and structure of this thesis
Chapter 2 – Study area: Garpenberg mine and Lappberget orebody
2.1 An introduction to Garpenberg mine
2.2 Geological setting and initial stress state
2.2.1 Lappberget orebody and weakness zones
2.2.2 Initial stress state and elastic rock mass properties
2.3 Mining method and sequencing
2.4 Geophysical and geotechnical monitoring in Lappberget
2.4.1 Extensometer data
2.4.2 Strain measurements
2.4.3 Microseismic data
2.5 Seismic activity and observed damage
Chapter 3 – Seismic data processing
3.1 Routines of seismic data acquisition and processing
3.1.1 Type of recorded seismic signals
3.1.2 Challenges and common errors in daily data processing
3.2 Picking consistency evaluation – The Wadati analysis
3.3 Evaluation of microseismic network performances
3.3.1 EMAP algorithm methodology
3.3.2 EMAP application to Lappberget microseismic network
3.4 Considerations about the extension of the analyzed area
3.5 Seismic source parameters estimation
3.5.1 Considerations on source parameters uncertainties
Chapter 4 – Rock mass response to mining
4.1 Spatiotemporal behavior of microseismic activity and mine blasts
4.1.1 Seismic sequences and clusters
4.2 Analysis of seismic source parameters
4.2.1 Temporal variation in b-value
4.3 What drives seismicity?
4.4 Analysis of geotechnical observations
4.5 Summary and discussion
Chapter 5 – Numerical modelling
5.1 Numerical modelling techniques
5.2 Model choice and strategy
5.3 Description of the model
5.3.1 Model geometry and boundaries
5.3.2 Model meshing
5.3.3 Initial and boundary conditions
5.3.4 Modelled elements and mechanical effect of paste fill
5.3.5 Constitutive laws and mechanical properties
5.3.6 Simulated mining sequence
5.4 Comparison with in situ geotechnical measurements
5.5 Model results and interpretations
5.5.1 Analysis of stress distribution
5.5.2 Analysis of strain distribution
5.5.3 Analysis of plastic zones and influence of weak geological materials
5.5.4 Temporal evolution of model parameters
5.6 Discussion and conclusion
Chapter 6 – Combined analysis of seismicity and numerical modelling
6.1 Relating induced seismicity with geomechanical modelling
6.2 Strategy of comparison in our work
6.3 Qualitative comparison at large-scale
6.3.1 Plastic zone and seismic activity
6.3.2 Instability criteria and seismic activity
6.4 Quantitative comparison at small-scale
6.4.1 Model and seismic parameters at punctual locations
6.4.2 Model and seismic parameters at spheres location
6.5 Summary and conclusion
Chapter 7 – Summary, conclusions and perspectives
7.1 Microseismic and geotechnical data analysis and interpretation
7.2 Numerical modelling and mining-induced seismicity
7.3 General perspectives