Screening of seismic records to perform time-history dynamic analyses of tailings dams: a power-spectral based approach
SScreening of seismic records for performing time-historydynamic analyses of tailings dams: a power-spectralbased approach
Nicols A. Labanda a,c,1 , Mauro G. Sottile a,b , Ignacio A. Cueto a,b , Alejo O.Sfriso a,b a SRK Consulting, Argentina. b Universidad de Buenos Aires. Facultad de Ingenier´ıa. Buenos Aires, Argentina. c Curtin University, School of Earth and Planetary Science, Bentley, Western Australia,Australia.
Abstract
Time-history deformation analyses of upstream-raised tailings dams useseismic records as input data. Such records must be representative of thein-situ seismicity in terms of a wide range of intensity measures (IMs)including PGA, Arias intensity, cumulative absolute velocity, source-to-sitedistance, duration, and others. No single IM is a sufficient descriptor ofseismic demand because different records, all of them compliant with any IM,can produce a very wide range of results from negligible damage to globalfailure. The use of brute-force, where hundreds of seismic records compliantwith a set of IMs is employed, has proven to be a reasonable workaround ofthis limitation, at least able to produce a probabilistic density function (PDF)of demand indicators like crest settlement of the dam or slope deformation.This procedure, however, requires a large number of numerical models to berun, and is therefore expensive and time-consuming. Brute-force analyses canbe optimized if an a-priori simple tool is used to predict which seismic recordswould yield a given demand, thus obtaining an estimate of the PDF of anydemand indicator with many less runs. In this study, a new semi-analyticalprocedure for evaluating the seismic demand imposed by a given seismicrecord on a tailings dam is proposed. The procedure employs the spectralproperties of the record filtered by those of the dam. Applications to damsunder strong earthquakes are presented and validated using numericalapproaches that show the robustness of the method by proving insensitivity toconstitutive models and mesh-independence. The proposed procedure is ableto produce an a-priori estimate of the damage potential of a given seismicrecord, thus reducing the number of runs required to produce a realisticdeformation analysis of tailings dams subjected to earthquakes and a robust ∗ Corresponding Author at: Curtin University, Bentley, Western Australia, Australia(nlabanda@fi.uba.ar).
Preprint submitted to Elsevier September 15, 2020 a r X i v : . [ phy s i c s . g e o - ph ] S e p stimate of the PDF of any given indicator of the seismic demand of the dam. Keywords:
Dynamic, Liquefaction, Tailing dams, Correlations, Time-historyanalysis
1. Introduction
Tailings are man-made materials created from mine-rock crushing, generallydeposited as a viscous mixture into storage facilities (TSFs). The lack of post-deposition compaction and the electrical interaction among the finer particlesentail loose states, which can be locked by early diagenesis cementation afterplacement [4]. The storage facility construction method can be downstream,centerline or upstream type; named after the crest movement direction duringthe raise. The latter is highly attractive from the economical point of view asit minimizes the construction volumes; however, they are the most unsafe andwith the most failure frequency, as stability largely relies on the strength of thetailings. Recent upstream-raised TSFs massive failures (such as Merriespruit,Mount Polley, Samarco and Brumadinho [32]) have depicted their vulnerabilityagainst liquefaction, having this kind of geotechnical structures an annual failureprobability five to ten times larger than hydroelectric dams [9].According to Ref. [30], liquefaction phenomenon is one of the main causesof tailings dams failures, where dynamic liquefaction represents around 15% ofthe cases. Liquefaction occurs when loose water-saturated tailings undergo anincrease of pore pressure and loss of strength due to undrained shearing or byinternal fabric collapse. In the context of the static liquefaction, due to thedifficulty of analyzing failure triggering events, updated internationalguidelines [1] recommend to conservatively assume that event will occur forbrittle/contractive near-saturated tailings. Thus, current design practiceinvolves limit equilibrium (LE) analyses adopting fully-softened shearstrengths; while safe, this simplistic approach makes no allowance for theamount of strain required for the material to start a strain-softening processleading to progressive failure, which is acceptable for designing new TSFs butfails short in assessing the risk posed by existing TSFs, both operating andabandoned.The scenario for dynamic liquefaction is totally different as dam codesrecommends static calculations for this assessment, considering post-seismicstrength in contractive saturated tailing, avoiding the inertia effects producedby an earthquake excitation. In this sense, numerical deformation analysesbecome imperative to better understand the dynamic liquefactionvulnerability and to warranty that the designed freeboards are adequate.One of the crucial aspects to perform dynamic liquefaction assessmentsrelies in the selection of a proper and most representative small subset ofseismic records from a usually huge data-base. During the developing ofearthquake engineering, extensive ground motion intensity parameters havebeen proposed to characterize the destructive potential of a record: peak2round velocity (PGV) and peak ground acceleration (PGA), the mostwidespread but limited intensity magnitudes [12]; Arias intensity, proportionalto the total energy content of the signal [2]; modified cumulative absolutevelocity, being the integral of the acceleration after application of a 5 cm/s acceleration threshold [25]; a normalized hysteretic energy, an empiricalrelation between dissipated shear energy and residual excess pore pressureratio [13], among many others proposals.In the context of liquefiable soils, new approaches has been studied anddiscussed in recently published papers. Kramer et al. [23] reviews proceduresto detect the time of liquefaction triggering, comparing their performance withempirical methods. The research is focused on signal analysis using short termFourier transform (STFT), spectrograms, wavelets transforms and Stockwellspectrum procedures, showing that the mean frequency content tends to reducein signals recorded above a liquefied stratum [24, 26, 40].Motivated by its simplistic and computationally efficient model, otherresearchers have used the Newmark’s model [28], to predict land-slides inslopes under shaking, adopting empirical modifications of the Arias intensityto characterize the excitation [8]. The combination between the Newmarkmodel, as a displacement estimator, and the Arias intensity has become,maybe, the most popular procedure for hazard analysis in geotechnics[19, 20, 3, 31, 10].Due to the improvement of computers capacity, in recent years severalcontributions have replaced analytical models of mechanical simulations withnumerical ones, mainly based on the finite element method. Ref. [27] hasattend the problem of tailing dam stability subjected to dynamic loads,putting efforts to detect resonance points by means of transfer functions[33, 16]. Jin et al. [21] have proposed a theoretical framework of mudslidesbased on experimental and numerical models, pointing out that instabilitymechanism of the tailings reservoir under seismic load are somehow the samecompared with results in bibliography, relating a proper design of a tailingdam to the seismic site condition rather than its constructive procedure. Its isworth to note that computational models, because on their versatility, is thestandard to perform forensic studies over real dam-breaks, and someremarkable works are recommended [17, 39, 38, 36].Despite all reached advances in this sense, the geotechnical industry and,particularly, the tailing engineering, is still using correlations of engineeringdemand parameters with classical intensity indicator, which usually lead tohighly scattered results [14]. This paper presents a novel and innovativeintensity measure for seismic record selection, stated as a generalization ofclassical intensity measures such as the Arias Intensity (AI), based on signalspectral properties. The proposal is validated using classical/analytical modelsand dynamic numerical methods to show its correlation with the induceddamage measured as the maximum displacement during the time-history. Forthe numerical modelling, the PM4Sand model proposed in Ref. [6] and [41], isused in order to simulate the pore pressure increase of tailings under cyclicloads, while the considered analytical model is the well-known Newmark3odel. Several seismic records are evaluated comparing our proposed intensitymagnitude with the classical, showing the improvement in resultingcorrelations and the novelty of our approach to select seismic records knowna-priori, which are will be the most dangerous.The paper is organized as follows: Some details and technical aspectsregarding the considered tailing dam and its composition are provided inSection 2. In Section 3 the constitutive model PM4S, used for liquefactionassessment, is calibrated from laboratory and field tests. In Section 4, theproposed methodology to an a-priori estimation of seismic liquefactionpotential is presented. Numerical results related to deformation and damageinduced by pore pressure accumulation and correlations with classical andproposed liquefaction indicators are presented and discussed in Section 5,where the predictive capability and efficiency of our approach is tested.Finally some conclusions and outlooks are drawn in Section 6.
2. Upstream tailing dam analized
The considered TSF facility is located in a high seismicity area where, in afirst stage, a 30-35 m height upstream-raised facility was built by dischargingtailings from the crest of two rock-fill starter dams and then raised usingdeposited beach tailings to form a slope profile. Several decades later, the TSFwas improved and brought back to operation, and continues to be raised withan upstream construction method, reaching a height of 60-65 m. An additionaltailings volume of approximately 3.4
M m , producing an rise of 10-15 m, isexpected to be deposited in the following years. The facility consists mainly ofa starter dam, aligned across a valley some 750 m apart; embankment raisesforming a 3H:1V slope; sandy silt/silty sand predominantly contractivetailings; vertical drains installed in the slope; and a reinforcement buttressbuilt at the toe. Figure 1: Geological units presented in the TSF.
The tailings have shown to be predominantly contractive, based oninterpretation of CPTu testing; in addition, the facility is built in a high4eismicity area; thus, the combination of these two aspects results in a highliquefaction risk, should tailings be both contractive and saturated. Anschematic cross section of the TSF in its final configuration is presented inFigure 1.
Several models has been carried out in order to estimate and quantify theTSF damage, fitting constitutive models and material parameters usinglaboratory and field test. In this section, a brief description of the generallayout of developed numerical models is presented.
Far field boundary conditions are used at the left and right ends of the model.At the bottom, a compliant base boundary condition is employed, and horizontalaccelerations-time signals are inputs used for the dynamic modeling. Accordingto the standard procedure for compliant bases, the input signal employed had50% of the acceleration of the recorded seismic record, to account for the factthat recorded signal is generated by the outcrop motion and composed by thesummation of an upward incoming and downward reflected waves. No verticalaccelerations are considered in this study.
To determine the Rayleigh damping coefficients, the fundamental frequencyof the tailings material is computed, as: f = V s,av H , (1)where v s,av is the average velocity of the soil unit and H is the height of theTSF. Considering a vertical cross-section on the tailings, at 150 m from the crestwith a height of 72 m to the mid-height, the confining pressure to calculate theaverage velocity is 200 kP a . The small strain stiffness can be estimated as: G = G ref (cid:18) pp ref (cid:19) m = 45 [ M P a ] (cid:18)
200 [ kP a ]100 [ kP a ] (cid:19) . = 75 [ M P a ] , (2)which translates into an average shear wave velocity of v s,av = 203 m/sec ;therefore, the natural frequency of the far-field tailings is approximately 0 . Hz .If the same calculation is done for a cross-section below the final crest: the heightis 74 m and the average mean pressure is 430 kP a ; which leads to the fundamentalfrequency of approximately 0 . Hz . Therefore, the expected range of tailingsfirst mode natural frequency is from 0 .
70 to 0 . Hz . If the same simplifiedcalculation is done for the buttress, the first mode natural frequency is estimatedto be approximately 2 . Hz , considering a height of 20 m and an average shear5ave velocity of 200 m/sec . Following [15], an upper bound target frequencyshould be defined as: f = f fund f , (3)where f fund is the fundamental frequency of the signal. For this dynamicanalysis, twenty-five signals are used; therefore, a representative value of 4 Hz is adopted for practical purposes. Then, the upper bound target frequency is5 Hz , when average tailings first mode natural frequency of 0 . Hz isconsidered. The chosen values for the Rayleigh coefficient α and β are 0 . . Figure 2 shows the construction procedure simulated in the presentednumerical models. The dynamic modeling stages start from the TSF finalconfiguration, where for the construction simulation a Hardening Soil Smallmodel is used due to the fact that PM4Sand usually fails to simulate staticscenarios. In addition, a good estimation of the effective pressure prior thedynamic simulation must be performed in order to estimate accurately therelative density D r needed to calibrate PM4Sand. After reaching the finalelevation, full consolidation of excess pore pressures is allowed, and tailingsmaterial is subsequently changed from HSS to PM4Sand following theprocedure described below:the average mean effective stress is obtained as an output for differentportions of the tailing material;the state parameter ψ is determined following the procedure presented in[35];the corresponding relative density is estimated based on the average meaneffective pressure and state parameter for each soil zone, following thecalibrated Critical State Line and the state parameter contours.Each tailing cluster ends up with the calibrated PM4Sand parameters,presented in the following section, with exception of the relative density whichis computed as a function of the average mean effective pressure at the end ofconstruction and the average expected state parameter. For sake of brevity,the rest of the geotechnical units will be not discussed in this paper due to itsredundancy for purposes of this new correlation.6 igure 2: Simulation of dam raise.
3. Constitutive model calibration for tailings
The PM4Sand constitutive model proposed by Ref. [41] is selected tosimulate the material behaviour in seismic loading and perform dynamicliquefaction assessments, calibrating model parameters from lab and field tests.Tailings samples recovery for lab tests was done by means of Mostapequipment, with locations adjacent to some CPT soundings. The laboratorytesting was done on tailings reconstituted samples, considering that theymight have been disturbed by sampling/handling/transportation/extrusionprocess; this hypothesis is completely valid for this type of tailings.General material characterization tests were performed on reconstitutedtailings samples, including: specific gravity, minimum and maximum drydensity, and particle size distribution. Samples has a composition of 40 % to77 % of Sand, 19 % to 50 % Silt and 4 % to 10 % of Clay size particles; theminimum and maximum dry densities are 1.24 to 1.32 t/m3 and 2.14 to 2.27t/m3, respectively; and the specific gravity is 2.75 to 2.79; this results in aminimum and maximum void ratio of 0.23-0.29 and 1.11-1.22, respectively.Two Cyclic Direct Simple Shear test (CDSS) are used for our dynamiccharacterization. The first has been performed at an initial vertical effectivestress of σ (cid:48) v = 200 kP a and an initial void ratio e = 0 . K = 0 .
65, the initial mean effective stress is approximately p (cid:48) ∼ = 150 kP a .Using a Cyclic Stress Ratio CSR=0.12, 30 cycles were needed to achievefailure. The second has been performed at an initial vertical effective stress of σ (cid:48) v = 200 kP a and an initial void ratio e = 0 . p (cid:48) − e state and its distance to theCritical State Line (CSL); it is defined as the difference between the currentvoid ratio and the void ratio at critical state for the current mean effectivestress, i.e. ψ = e − e c . Correlations and methods have been developed andused to estimate state parameter from CPTu measurements ψ = f [ Q p , B q k, m ]; however, the adjustment of k and m coefficients weremainly elaborated for sand-like materials, and might not be fullyrepresentative of silt-like tailings. Comparisons with laboratory results ofdifferent CSLs for both, NorSand and PM4Sand, are presented in Figure 3(a)in terms of relative density D r and effective pressure p (cid:48) and Figure 3(b) interms of void ratio e and the effective pressure. It can be seen that the best fitof the state parameter with the laboratory test is obtained for Γ ∼ = 1. Forfurther details in this regards, the reader is referred to read [35], where thespatial distribution obtained with this methodology is presented in Figure 4.The CSL is computed as e CSL = Γ − λ · log p (cid:48) , (4)considering parameters presented in Table 1. Table 1: Critical state line parameters.
Sample Γ λ e min e max CS Data N ◦ N ◦ ψ as8 Mean Effective Pressure p'[kPa] R e l a t i v e D en s i t y D r [ - ] NorSand = -0.15PM4Sand = -0.15NorSand = -0.05PM4Sand = -0.05NorSand = 0.05PM4Sand = 0.05NorSand = 0.15PM4Sand = 0.15NorSand = 0.25PM4Sand = 0.25NorSand - CSLPM4Sand - CSLCS Data N°1CS Data N°2 (a) Relative density D r versus mean effective pressure p (cid:48) plot. Mean Effective Pressure p'[kPa] V o i d R a t i o e [ - ] NorSand - CSLPM4Sand - CSLCS Data N°1CS Data N°2 (b) Void Ratio e versus mean effective pressure p (cid:48) plot.Figure 3: Critical State Line (CSL) calibration and comparisons with isotropicallyconsolidated undrained and drained tests. D r = e max − ( e CSL + ψ ) e max − e min , (5)where the void ratio in the critical state line e CSL is computed considering theeffective pressure p (cid:48) obtained at the end of the dam raising, after consolidation.Using the state parameters distributions presented in Figure 4, the resultingdistribution of relative densities used as an input in our PM4Sand model ispresented in Figure 5. Figure 4: Distribution of the state parameter ψ within the dam.Figure 5: Distribution of relative density D r considered in seismic simulations. In order to validate our calibrations, comparisons with two cyclic directshear test are presented in Figure 6, where a reasonably good fitting isobtained with the proposed model. In Table 2 are summarized all parametersfor Hardening soil small used for the dam raise, and in Table 3 PM4Sand’sparameters used for the dynamic liquefaction analysis. For sake of brevity, thediscussion about the HSsmall calibration will be avoided and left to bediscussed in future contributions. 10
Cycle P o r e W a t e r P r e ss u r e R a t i o [ % ] Lab testPM4Sand
Cycle -10-50510 S hea r S t r a i n [ % ] Lab testPM4Sand
Vertical Stress Ratio p'/p' -0.15-0.1-0.0500.050.10.15 C yc li c S t r e ss R a t i o / p ' Lab testPM4Sand -10 -5 0 5 10
Shear Strain [%] -0.15-0.1-0.0500.050.10.15 C yc li c S t r e ss R a t i o / p ' Lab testPM4Sand (a) PM4Sand vs. Laboratory tests (N ◦ Cycle P o r e W a t e r P r e ss u r e R a t i o [ % ] Lab testPM4Sand
Cycle -6-4-20246 S hea r S t r a i n [ % ] Lab testPM4Sand
Vertical Stress Ratio p'/p' -0.15-0.1-0.0500.050.10.15 C yc li c S t r e ss R a t i o / p ' Lab testPM4Sand -6 -4 -2 0 2 4 6 8
Shear Strain [%] -0.15-0.1-0.0500.050.10.15 C yc li c S t r e ss R a t i o / p ' Lab testPM4Sand (b) PM4Sand vs. Laboratory tests (N ◦ able 2: HSsmall parameters for tailing raise used in numerical simulations. Unit HSsmall γ sat kN/m φ (cid:48) ◦ c (cid:48) kP a ψ ◦ G ref M P a γ . - 10 − E refur M P a E ref M P a E refoed M P a m - 0.75 ν ur - 0.20OCR - 1.00 K nc - 0.50 Table 3: PM4Sand parameters for tailings used in dynamic liquefaction numerical simulations.
Unit Parameters γ sat kN/m G - 450 h p - 0.75 p ref kP a e max - 1.166 e min - 0.257 n b - 0.50 n d - 0.10 φ (cid:48) cv ◦ ν - 0.20 Q - 11.8 R - 3.4 k − m/s 112 . A-priori liquefaction risk estimate of a seismic record database This section represents the core of the paper where, first, the proposedintensity measure is described and mathematically defined and then,calculations are performed to a set of seismic signals obtaining the powerspectral based intensity measure for each earthquake.
Calculations to obtain the spectral power content of a seismic signal arebased on the Fast Fourier Transform (FFT), an efficient implementation ofthe discrete Fourier transform (DFT). This paper will be stated considering adiscrete space avoiding algorithm issues, in order to explain clearly the physicalaspects.Let { a n } = a , a , ..., a N − be a finite set of N elements uniformly spacedof time-history accelerations, the DFT is defined by means of Euler’s formula: F { a n } ( k ) = { A k } = N − (cid:88) n =0 a n · e − i πk nN = N − (cid:88) n =0 a n · (cid:20) cos (cid:18) πN kn (cid:19) + i · sin (cid:18) πN kn (cid:19)(cid:21) , (6)where { A k } is a set of complex vectors which represents the amplitude andphase of a complex sinusoidal component and k an integer representing thefrequency domain.The power spectrum density in terms of the frequency is defined as S xx ( k ) = (cid:107) F { a n } ( k ) (cid:107) , (7)while the total spectral power of the signal is expresed as P −∞ = ∞ (cid:88) k =0 S xx ( k ) ∆ k, (8)with ∆ k the frequency sampling. In this paper, the seismic intensity measureused is the spectral power expressed in equation (8), where the final frequencyconsidered for our calculations is the limit of power accumulation i.e., P − XHz is the accumulated power between the frequencies 0 to
XHz . It is worth to notethat, due to Parseval’s Theorem N − (cid:88) n =0 (cid:107) a n (cid:107) = ∞ (cid:88) k =0 (cid:107) F ( k ) (cid:107) , (9)which means that, when the considered power spectrum is computedconsidering all the frequency domain, the intensity measure expressed in termsof the spectral decomposition tends to represents the same quantity13epresented by classical intensity measures, based on the integration of theseismic signal such as the Arias intensity. In this way, our proposal is stated asa generalization of these classical intensity measures.Finally, the spectrogram expressing a signal decomposition in terms of time,frequency and spectral power, is plotted in terms of the spectral power expressedin decibels dB , computed as P dB = 10 log (cid:18) PP r (cid:19) , (10)where P is the computed spectral power and P r = 10 . is a reference power.The reference power behaves like a simple shift in the accumulated power anddoes not modify the proposed correlation. In order to evaluate the liquefaction risk of the tailing dam, a set of 25seismic records are selected based on the 50th percentile of the 7500-year event,for which a peak ground acceleration (PGA) of 0.78 g is expected. Their maincharacteristics are summarized in Table 4 and the signals are included in FigureA.19, where a Hamming window is used for the calculations. Among the 25seismic records employed, there are 20 that correspond to seismographs locatedon dense soil/soft rock (NEHRP site class C); two (records 3 and 8) belong tofirm/hard rock (NEHRP site class B), and three (records 4, 7 and 23) belongto stiff soils (NEHRP site class D). For each seismic record, the acceleration-time signal and spectrogram, is computed by using short-time fourier transform[29] and presented in Figure A.19. Our new intensity measure is included inthe table together with classical intensity measures like arias intensity (AI),Cumulative Accelerate Value (CAV) and Cumulative Accelerate Value above0 . g (CAV5), accumulating the spectral power between 0 Hz and 2 Hz , andexpressed in relative terms, i.e. the argument of the logarithm, as was explainedin equation (10).Figure 7 (a) plots the cumulative spectral power of considered seismic signals.It can be seen that some of the signal have more power accumulated at lowfrequencies and, while the frequency windows increases, others increases theirspectral power for high frequencies. The curves has been ordered such that,those with higher spectral power in low frequencies are plotted in red colors,while those with less power are printed in blue. In this sense, seismic recordsnumber 8, 4, 19 and 7 are those with more power from 0 to 2 Hz, while 18, 3, 6and 1 are the ones with less. Another way to express these results is presentedin Figure 7 (b) where a mobile Hamming window with 2 Hz width is used. It canbe seen that some signals like number 9 has its power accumulated in mediumfrequencies, while its content for low and high is relatively low and, others likenumber 8, has its major proportion spectral power in the low band, while inmedium and high is progressively negligible.In the following section, it will be demonstrated that the destructivepotential of a seismic records in tailing dams is strictly related with thespectral power content at low frequencies.14 Frequency [Hz] C u m u l a t i v e s pe c t r a l po w e r Low frequencies Medium frequencies High frequencies (a) Spectral power calculated using a Hamming windows fixed from0 Hz . Frequency [Hz] S pe c t r a l po w e r [ d B ] - W i ndo w w i d t h . [ H z ] (b) Spectral power calculated using a Hamming windows with 2 Hz width.Figure 7: Acumulated spectral power for the considered set of signals, obtained withsprectrograms presented in Figure A.21. a b l e : S e i s m i c r ec o r d s u s e d f o r t h e s e i s m i c a n a l y s i s , s c a l e d t o P G A o f . . I D E v e n t N a m e R e c o r dS t a t i o n D u r a t i o n A I C AV C AV P − . H z P r P − . H z P r [ s e c ][ mm / s e c ][ mm / s e c ][ mm / s e c ][ − ][ − ] N a h a nn i C a n a d a S i t e . .
58 2 D u zc e T u r k e y L a m o n t . .
67 3 L a nd e r s L u c e r n e . .
88 4 K o b e J a p a n K J M A . .
10 5 S a nS a l v a d o r G e o t e c h I n v C . . .
66 6 P a r k fi e l d - C A P a r k fi e l d - C h o l a m e E . .
42 7 C hu e t s u - o k i T a m a t i Y o n e I z u m o z a k i . .
05 8 T a b a s I r a n T a b a s . .
07 9 I w a t e M Y G . .
75 10 N o r t h r i d g e - J e n s e n F il t e r P l a n t B u il d i n g294947138871416646433 . .
35 11 C h i - C h i T a i w a n TC U . .
90 12 M a mm o t h L a k e s - L o n g V a ll e y D a m . .
46 13 L o m a P r i e t a W AH O . .
66 14 V i c t o r i a M e x i c o C e rr o P r i e t o253579128681304531633 . .
50 15 L o m a P r i e t a C o rr a li t o s . .
24 16 C h i - C h i T a i w a n C HY . .
29 17 C oa li n ga - T r a n s m i tt e r H ill . .
04 18 C hu e t s u - o k i J o e t s u O s h i m a k u O k a60303210029108653412 . .
03 19 C . M e nd o c i n o P e t r o li a366556181851889987179 . .
93 20 C oa li n ga - O il C i t y . .
23 21 C h i - C h i T a i w a n C HY . .
82 22 C h r i s t c hu r c h N Z L P CC . .
39 23 N / A W a r d F i r e S t . .
22 24 N / AAN G O L . .
94 25 N / A P I C A . .
16s was explained in equation (9), the power-spectral approach is ageneralization of classical seismic intensity measures, performing a filtering ofthe power in certain frequencies that just brings scatter in correlations. It willbe shown that these spurious powers in the case of dams and tailings are thoseat low and medium-low frequencies (less than 5 Hz ). Figure 8 shows thecomparisons AI in Fig. 8 (a) and (b), showing that Parseval’s theorem isaccomplished. Fig. 8 (c) to (f) shows CAV and CAV5 with cumulativespectral power within different ranges of frequencies. When the bandfrequency is narrow, i.e. up to 2.0 Hz, a considerable scatter is between theintensity measures in all cases but, as the frequency band becomes wider, theaccumulated spectral power tends to represents AI, CAV and CAV5 obtaininga perfect fitting in all cases between intensity measures. It is worth to notethat, for some records like 21 or 24, the filtering produced by the intensitymeasure’s definition for CAV and CAV5, does not represents a straightenergetic generalization.
5. Numerical results
Several results are presented in this section in order to validate thepower-spectral based intensity parameter. Three approaches has been selectedto quantify the damage level: the well-known Newmark displacement modelused to obtain rough estimations of displacements in dams under earthquakeexcitation, the finite element numerical model using a constitutive model intailing to take into account the dynamic liquefaction of these geothecnicalstructures under cyclic load, and an extra set of models to validate theproposed approach with one of the most constitutive models used in industryi.e. Hardening Soil Small.
First, the proposed intensity measure is contrasted with the Newmarkdisplacement [28] for being a popular damage indicator in bibliography[11, 37, 22, 7]. Parameters presented in Table 2 have been used to obtain thedisplacements in all cases.Results are plotted, comparing Newmark displacement with Arias Intensityin Fig. 9 (a), CAV in Fig. 9 (b), CAV5 in Fig. 9 (c), cumulative spectral powerbetween 0 and 1.0 Hz P − . Hz in Fig.9 (d), P − . Hz in Fig. 9 (e) and finally P − . Hz in Fig. 9 (f).With classical intensity measures, a R = 0 .
60 is obtained for the best case(AI) in the linear regression with the maximum time-history displacement, beinga poor strategy to know a-priori the most devastating seismic record for theconsidered structure. Results improves considerably for the proposed intensitymeasure, where the fitting arises to R = 0 .
81 when the spectral power isaccumulated between 0 and 5.0 Hz, suggesting that the most dangerous powercontents are placed in low and medium to low frequencies.17
Power between 0 and 2.0 Hz A r i a s I n t en s i t y Correlation power and Arias Intensity (a) Arias intensity versus spectral powerbetween 0 and 2.0 Hz
Power between 0 and 15.0 Hz A r i a s I n t en s i t y Correlation power and Arias Intensity (b) Arias intensity versus spectral powerbetween 0 and 15.0 Hz
Power between 0 and 2.0 Hz C AV [ mm / s e c ] Correlation power and CAV (c) Cumulative Absolute Velocity versusspectral power between 0 and 2.0 Hz
Power between 0 and 15.0 Hz C AV [ mm / s e c ] Correlation power and CAV (d) Cumulative Absolute Velocity versusspectral power between 0 and 15.0 Hz
Power between 0 and 2.0 Hz C AV [ mm / s e c ] Correlation power and CAV5 (e) Cumulative Absolute Velocity above0.05g versus spectral power between 0 and2.0 Hz
Power between 0 and 15.0 Hz C AV [ mm / s e c ] Correlation power and CAV5 (f) Cumulative Absolute Velocity above0.05g versus spectral power between 0 and15.0 HzFigure 8: Comparisons of the proposed intensity measure based on spectral power and classicalmeasures. Arias Intensity [mm/sec] N e w m a r k d i s p l a c e m en t [ m ] Correlation between Newmark displacement and Arias Intensity R = 0.60454 (a) Newmark displacement versus AriasIntensity CAV [mm/sec] N e w m a r k d i s p l a c e m en t [ m ] Correlation between Newmark displacement and CAV R = 0.44552 (b) Newmark displacement versusCumulative Absolute Velocity. CAV5 [mm/sec] N e w m a r k d i s p l a c e m en t [ m ] Correlation between Newmark displacement and CAV5 R = 0.44596 (c) Newmark displacement versusCumulative Absolute Velocity above0 . g . Power between 0 and 1.0 Hz N e w m a r k d i s p l a c e m en t [ m ] Correlation between Newmark displacement and power R = 0.47827 (d) Newmark displacement versusspectral power between 0 and 1.0 Hz. Power between 0 and 2.0 Hz N e w m a r k d i s p l a c e m en t [ m ] Correlation between Newmark displacement and power R = 0.54618 (e) Newmark displacement versusspectral power between 0 and 2.0 Hz. Power between 0 and 5.0 Hz N e w m a r k d i s p l a c e m en t [ m ] Correlation between Newmark displacement and power R = 0.81536 (f) Newmark displacement versusspectral power between 0 and 5.0 Hz.Figure 9: Newmark displacement compared with classical and the proposed intensity measure. igure 10: Selected point to measure the displacements in the buttress. Several finite element simulations have been carried out using the describedcalibration of PM4Sand, using as input signal all seismic records presented inprevious section. Two points where selected to quantify the damage induced bythe earthquake, one in the crest and one in the base or toe of the buttress (seeFigure 10).Figure 11 shows the map displacement of the tailing dam in its buttress zonefor each considered seismic record. The results have been ordered by those withsmall (SD), moderate (MD) and large displacements (LD). The results showthat in some cases like records 1 to 3, damage is negligible with displacementsconcentrated over the edge of the buttress. Results obtained for records 4, 8, 16,19, 21 among others, demonstrate a considerable damage with a huge portionof tailings sliding down due to the liquefaction.Figure 17 (a) to (c) shows the time history of the base displacements, whileFigure 17 (d) to (f) shows the time history for the crest point. When resultsare analysed in this sense, it seems like there is no relation, obtaining a greatscatter between all simulations. The results varies from 0.1 to 4.0 meters forthe base point, while values ranging from 0.2 to 8.0 meters have been obtainedfor the crest displacement.Some attention must be given to the results of seismic records N ◦ ◦ ◦ ◦ ◦ ◦
4. This phenomenon is clearlyinterpreted by the intensity measure expressed in terms of the power spectrum,being the power concentrated in low frequencies higher in one case compared tothe other and, consequently, a larger damage is induced.Figure 15 shows the maximum time-history displacement obtained inprevious simulations and different seismic intensity measures. Similarly to the20 igure 11: Map of displacements obtained for each considered seismic record. Time[sec] D i s p l a c e m en t | U | [ m ] Results in Base node for Green zone
Seismic Record N°1Seismic Record N°2Seismic Record N°3Seismic Record N°6Seismic Record N°9Seismic Record N°12Seismic Record N°13Seismic Record N°14Seismic Record N°17Seismic Record N°18Seismic Record N°20Seismic Record N°22Seismic Record N°23Seismic Record N°24Seismic Record N°25 (a) Small displacements - Base point
Time[sec] D i s p l a c e m en t | U | [ m ] Results in Base node for Yellow zone
Seismic Record N°5Seismic Record N°10Seismic Record N°15 (b) Moderate displacements - Base point
Time[sec] D i s p l a c e m en t | U | [ m ] Results in Base node for Red zone
Seismic Record N°4Seismic Record N°7Seismic Record N°8Seismic Record N°11Seismic Record N°16Seismic Record N°19Seismic Record N°21 (c) Large displacements - Base point
Time[sec] D i s p l a c e m en t | U | [ m ] Results in Crest node for Green zone
Seismic Record N°1Seismic Record N°2Seismic Record N°3Seismic Record N°6Seismic Record N°9Seismic Record N°12Seismic Record N°13Seismic Record N°14Seismic Record N°17Seismic Record N°18Seismic Record N°20Seismic Record N°22Seismic Record N°23Seismic Record N°24Seismic Record N°25 (d) Small displacements - Crest point
Time[sec] D i s p l a c e m en t | U | [ m ] Results in Crest node for Yellow zone
Seismic Record N°5Seismic Record N°10Seismic Record N°15 (e) Moderate displacements - Crest point
Time[sec] D i s p l a c e m en t | U | [ m ] Results in Crest node for Red zone
Seismic Record N°4Seismic Record N°7Seismic Record N°8Seismic Record N°11Seismic Record N°16Seismic Record N°19Seismic Record N°21 (f) Large displacements - Crest pointFigure 12: Displacements history for base and crest point. Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°9
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (a) Record N ◦ AI = 12444 and P − Hz /P r = 28378 . Time [Sec] -10-505 A cc e l . [ m / s ] Seismic record N°4
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (b) Record N ◦ AI = 9057 and P − Hz /P r = 135265 . ◦
9. (b) Global failure for seismic record N ◦ R = 0 .
07 in Figure 15 (a), while for CAV and CAV5 in Fig. 15(b) and (c) respectively, the correlation is even worst. The spectral powerintensity measure presented in Fig. 15 (d) shows an accurate fitting with a R = 0 .
95 when low frequencies are considered, i.e. accumulated spectralpower between 0 and 2.0 Hz P − . Hz . When the frequency band becomeswider, the scatter increases obtaining a similar pattern than the one obtainedfor the AI (Fig. 15 (e)). A new set of finite element models have been performed by consideringthe most used constitutive model in the industry, the Hardening Soil Model(HSsmall) in order to assess the proposal from another point of view. The sameparameters used for the dam raising have been used in this case, and the sameseismic records are considered.Figure 16 shows the results of the considered seismic intensity measures,compared with the crest displacement obtained for the present case. Similarlyto the PM4Sand case, the correlation obtained for classical intensity measuresare rather poor. The best fit is also obtained by the AI with a R = 0 .
19 inFig. 16 (a), much better than the previous case. Results for CAV and CAV5are presented in Fig. 16(b) and (c) respectively. When the spectral power isconsidered the correlation becomes much better, although with a lower R thanthe one obtained when the PM4Sand model is considered. For our proposal,the best fit is obtained when the spectral power is accumulated between 0 and2 . Hz , where the R = 0 .
86 as shown in Fig. 16 (e), obtaining the sameconclusion where low and medium-low frequencies are the most dangerous andthe ones that provokes the dam failure.When maximum time-history base displacements are compared with theintensity measures, similar results are obtained. Figure 17 (a) to (c) show thesedisplacement compared with AI, CAV and CAV5 respectively, showing a bestfitting for the AI with an R = 0 .
22. Figure 17 (d) shows the results comparingwith the spectral power between 0 and 2.5 Hz, obtaining a highly accuratecorrelation.
6. Conclusions
The dynamic numerical analyses show that the type of failure and itsassociated residual displacements are highly dependent on the power of theearthquakes at low and medium-low frequencies. While all of the 25 groundmotions employed have the energy and acceleration of a Maximum DesignEarthquake, only 8 of them induce a global failure and 17 entail no failure orlocal failure with minor damages to the considered tailing dam whenPM4Sand model is considered in the calculations. For the 17 no-failurescenarios, the increment in displacements is negligible, and for the 8 ground24
Arias Intensity [mm/sec] c r e s t m a x i m u m d i s p l a c e m en t [ m ] Correlation between crest maximum displacement and Arias Intensity R = 0.079099 (a) Crest displacements versus AriasIntensity CAV [mm/sec] c r e s t m a x i m u m d i s p l a c e m en t [ m ] Correlation between crest maximum displacement and CAV R = 0.053047 (b) Crest displacements versusCumulative Absolute Velocity. CAV5 [mm/sec] c r e s t m a x i m u m d i s p l a c e m en t [ m ] Correlation between crest maximum displacement and CAV5 R = 0.053844 (c) Crest displacements versusCumulative Absolute Velocity above0 . g . (d) Crest displacements versus powerbetween 0 and 2.0 Hz.(e) Crest displacements versus powerbetween 0 and 15.0 Hz.Figure 15: Crest displacements obtained with the PM4Sand model compared with classicaland the proposed intensity measure. Arias Intensity [mm/sec] c r e s t m a x i m u m d i s p l a c e m en t [ m ] Correlation between crest maximum displacement and Arias Intensity R = 0.19678 (a) Crest displacements versus AriasIntensity CAV [mm/sec] c r e s t m a x i m u m d i s p l a c e m en t [ m ] Correlation between crest maximum displacement and CAV R = 0.13628 (b) Crest displacements versusCumulative Absolute Velocity. CAV5 [mm/sec] c r e s t m a x i m u m d i s p l a c e m en t [ m ] Correlation between crest maximum displacement and CAV5 R = 0.13293 (c) Crest displacements versusCumulative Absolute Velocity above0 . g . Power between 0 and 1.0 Hz c r e s t m a x i m u m d i s p l a c e m en t [ m ] Correlation between crest maximum displacement and power R = 0.83699 (d) Crest displacements versus powerbetween 0 and 1.0 Hz. Power between 0 and 2.5 Hz c r e s t m a x i m u m d i s p l a c e m en t [ m ] Correlation between crest maximum displacement and power R = 0.86261 (e) Crest displacements versus powerbetween 0 and 2.5 Hz. Power between 0 and 15.0 Hz c r e s t m a x i m u m d i s p l a c e m en t [ m ] Correlation between crest maximum displacement and power R = 0.20964 (f) Crest displacements versus powerbetween 0 and 15.0 Hz.Figure 16: Crest displacements obtained with the Hardening Soil Small (HSsmall) modelcompared with classical and the proposed intensity measure. Arias Intensity [mm/sec] B a s e m a x i m u m d i s p l a c e m en t [ m ] Correlation between Base maximum displacement and Arias Intensity R = 0.22615 (a) Base displacements versus AriasIntensity. CAV [mm/sec] B a s e m a x i m u m d i s p l a c e m en t [ m ] Correlation between Base maximum displacement and CAV R = 0.17865 (b) Base displacements versusCumulative Absolute Velocity. CAV5 [mm/sec] B a s e m a x i m u m d i s p l a c e m en t [ m ] Correlation between Base maximum displacement and CAV5 R = 0.17653 (c) Base displacements versus CumulativeAbsolute Velocity above 0 . g . Power between 0 and 2.5 Hz B a s e m a x i m u m d i s p l a c e m en t [ m ] Correlation between Base maximum displacement and power R = 0.91381 (d) Base displacements versus powerbetween 0 and 2.5 Hz.Figure 17: Base displacements obtained with the Hardening Soil Small model compared withclassical and the proposed intensity measure. motions inducing failure, the difference is small. It must be noted that fromthe 8 records that induce significant displacements, two were recorded on stiffsoil, which might be overly conservative for this site.It has been shown that the proposed method for seismic selection based onthe spectral power content has a good correlation with both, numerical andanalytical models, being a generalization of other classical and well-knownintensity parameter such as Arias Intensity, Cumulative Absolute Velocity andCumulative Absolute Velocity above 0 . g . Our results demonstrates that thekey aspect to take into account is the filtering of spectral power in thefrequency band which are in resonance with the structures. When the classicalNewmark model is considered to calculate the maximum failure displacementof the dam, a reasonable correlation in obtained for spectral powers within the0 to 5 Hz range, with a R = 0 .
81. Even for the worst results, the proposal isstill are a really accurate correlation. 27he best correlations have been obtained with the numerical models. Twoscenarios have been considered, the first where a constitutive model to capturedynamic liquefaction (PM4Sand) is used, and the second where the sameconstitutive model used to raise the dam (Hardening Soil Small) has beenused. For these cases, the resulting outcomes have been highly accurate. Thecorrelations obtained for Hardening soil model compares the damage,measured as the displacement of the crest, versus the spectral power, showinga good fitting between both obtaining a R = 0 .
86 when the power content iscomputed between 0 and 2 . Hz . On the other hand, it can be seen thatclassical intensity measures like cumulative absolute velocity and Ariasintensity have a poor correlation with factors R lower than 0 .
15. When thedamage measure is done by means of the base displacement, good correlationsare also obtained.If dynamic liquefaction is considered using PM4Sand model, correlationsobtained are almost perfect with R = 0 .
95 for a cumulative spectral powerbetween 0 and 2 . Hz . While the frequency windows increases, the fittingbetween damage and the intensity measure starts to diverge. Similarly to theHardening soil small case, the worst correlations have been obtained for thecumulative absolute velocity and the Arias intensity.Some plots comparing our proposal for intensity measures and classicalintensity measures have been also presented, showing that the spectral powerapproach is just a power filtering of classical approaches, being a generalizationof other well-known approaches. Summarizing, the paper has shown that thespectral power criteria can be an accurate approach to select a reduced set ofseismic signals within a huge database, being a promising criteria to be used toassess damage using numerical and/or analytical methods. References [1] ANCOLD, editor.
Australian National committee on Large Dams.Guidelines on Tailings Dams. Planning, De-sign, Construction, Operationand Closure.
Australian National Committee on Large Dams, 2019.[2] A. Arias. A measure of earthqueake intensity.
Seismic design for nuclearpower plants. Cambridge, Massachusetts Institute of Technology Press ,pages 438–483, 1990.[3] R. J. Armstrong, E. J. Malvick, H. Hansra, and C. Eiermann.
Evaluationof Empirical Predictive Models Used to Predict Earthquake-Induced SlopeDeformations , pages 1250–1259.[4] R. C. Bachus, M. Terzariol, C. Pasten, S. H. Chong, S. Dai, M. S. Cha,S. Kim, J. Jang, E. Papadopoulos, S. Roshankhah, L. Lei, A. Garcia,J. Park, A. Sivaram, F. Santamarina, X. Ren, and J. C. Santamarina.Characterization and engineering properties of dry and ponded class-f fly ash.
Journal of Geotechnical and Geoenvironmental Engineering ,145(3):04019003, 2019. 285] K. Been and M. G. Jefferies. A state parameter for sands.
Gotechnique ,35(2):99–112, 1985.[6] R. Boulanger and K. Ziotopoulou. Formulation of a sand plasticity plane-strain model for earthquake engineering applications.
Soil Dynamics andEarthquake Engineering , 53:254 – 267, 2013.[7] E. Cattoni, D. Salciarini, and C. Tamagnini. A generalized newmarkmethod for the assessment of permanent displacements of flexible retainingstructures under seismic loading conditions.
Soil Dynamics and EarthquakeEngineering , 117:221 – 233, 2019.[8] K. Chousianitis, V. D. Gaudio, I. Kalogeras, and A. Ganas. Predictivemodel of arias intensity and newmark displacement for regional scaleevaluation of earthquake-induced landslide hazard in greece.
Soil Dynamicsand Earthquake Engineering , 65:11 – 29, 2014.[9] M. P. Davies. Tailings impoundment failures: Are geotechnical engineerslistening?
Geotechnical News , 20:31 – 36, 2002.[10] M. Deyanova, C. G. Lai, and M. Martinelli. Displacement based parametricstudy on the seismic response of gravity earth-retaining walls.
SoilDynamics and Earthquake Engineering , 80:210 – 224, 2016.[11] W. Du, D. Huang, and G. Wang. Quantification of model uncertaintyand variability in newmark displacement analysis.
Soil Dynamics andEarthquake Engineering , 109:286 – 298, 2018.[12] B. Ebrahimian. Non-linear numerical analysis of earthquake- induceddeformation of earth-fill dams.
Advances in Geotechnical EarthquakeEngineering - Soil Liquefaction and Seismic Safety of Dams andMonuments. Edited by Abbas Moustafa, Intech Publisher , pages 1–10, 2012.[13] R. Green, J. Mitchell, and C. Polito. An energy-based excess pore pressuregeneration model for cohesionless soils.
Proceedings of the John BookerMemorial Symposium, Sydney, New South Wales, Australia. A.A. BalkemaPublishers, Rotterdam, Netherlands , pages 1–9, 2000.[14] M. A. Hariri-Ardebili and J. Xu. Efficient seismic reliability analysis oflarge-scale coupled systems including epistemic and aleatory uncertainties.
Soil Dynamics and Earthquake Engineering , 116:761 – 773, 2019.[15] M. Hudson, I. Idriss, and M. Bourke. Quad4m user’s manual. 1994.[16] J.-H. Hwang, C.-P. Wu, and S.-C. Wang. Seismic record analysis of theliyutan earth dam.
Canadian Geotechnical Journal , 44(11):1351–1377,2007. 2917] K. Ishihara, K. Ueno, S. Yamada, S. Yasuda, and T. Yoneoka. Breachof a tailings dam in the 2011 earthquake in japan.
Soil Dynamics andEarthquake Engineering , 68:3 – 22, 2015. The Kenji Ishihara Lecture SeriesInaugural Articles.[18] M. G. Jefferies. Nor-sand: a simle critical state model for sand.
Gotechnique , 43(1):91–103, 1993.[19] R. Jibson and D. Keefer. Analysis of the seismic origin of landslides:examples from the new madrid seismic zone.
Geological Society of AmericaBulletin , 105(4):521–536, 1993.[20] R. W. Jibson. Regression models for estimating coseismic landslidedisplacement.
Engineering Geology , 91(2):209 – 218, 2007.[21] J. Jin, C. Song, B. Liang, Y. Chen, and M. Su. Dynamic characteristicsof tailings reservoir under seismic load.
Environmental Earth Sciences ,77(18):654, Sep 2018.[22] T. Kokusho. Energy-based newmark method for earthquake-induced slopedisplacements.
Soil Dynamics and Earthquake Engineering , 121:121 – 134,2019.[23] S. Kramer, S. Sideras, and M. Greenfield. The timing of liquefaction andits utility in liquefaction hazard evaluation.
Soil Dynamics and EarthquakeEngineering , 91:133 – 146, 2016. 6ICEGE Earthquake GeotechincalEngineeering.[24] S. Kramer, S. Sideras, M. Greenfield, and B. Hushmand. Liquefaction,ground motions, and pore pressures at the wildlife liquefaction array in the1987 superstition hills earthquake. volume 2018-June, pages 384–402, 2018.[25] S. L. Kramer and R. A. Mitchell. Ground motion intensity measures forliquefaction hazard evaluation.
Earthquake Spectra , 22(2):413–438, 2006.[26] K. C. Meza-Fajardo, C. Varone, L. Lenti, S. Martino, and J.-F. Semblat.Surface wave quantification in a highly heterogeneous alluvial basin: Casestudy of the fosso di vallerano valley, rome, italy.
Soil Dynamics andEarthquake Engineering , 120:292 – 300, 2019.[27] M. Naeini and A. Akhtarpour. Numerical analysis of seismic stability of ahigh centerline tailings dam.
Soil Dynamics and Earthquake Engineering ,107:179 – 194, 2018.[28] N. M. Newmark. Effects of earthquakes on dams and embankments.
Gotechnique , 15(2):139–160, 1965.[29] A. V. Oppenheim, R. W. Schafer, and J. R. Buck.
Discrete-Time SignalProcessing . 1999. 3030] M. Rico, G. Benito, A. Salgueiro, A. Dez-Herrero, and H. Pereira. Reportedtailings dam failures: A review of the european incidents in the worldwidecontext.
Journal of Hazardous Materials , 152(2):846 – 852, 2008.[31] R. Roy, D. Ghosh, and G. Bhattacharya. Influence of strong motioncharacteristics on permanent displacement of slopes.
Landslides , 13(2):279–292, Apr 2016.[32] J. C. Santamarina, L. A. Torres-Cruz, and R. C. Bachus. Why coal ashand tailings dam disasters occur.
Science , 364(6440):526–528, 2019.[33] R. T. Severn. Dynamic behaviour of arch dams. In J. O. Pedro, editor,
Arch Dams , pages 289–403, Vienna, 1999. Springer Vienna.[34] D. Shuttle and M. Jefferies. Determining silt state from cptu.
GeotechnicalResearch , 3(3):90–118, 2016.[35] M. G. Sottile, A. Kerguelen, and S. A. O. A comparison of procedures fordetermining the state parameter of silt-like tailings. In P. D. Kalumba,editor,
The 17th African Regional Conference on Soil Mechanics andGeotechnical Engineering , pages 1–5, Cape Town, South Africa, 2019.[36] W. ´Swidzi´nski, A. Korzec, and K. Wo´zniczko.
Stability Analysis of ˙ZelaznyMost Tailings Dam Loaded by Mining-Induced Earthquakes , pages 303–311.Springer International Publishing, Cham, 2016.[37] G. Veylon, L.-H. Luu, S. Merckl, P.-Y. Bard, A. Delvalle, C. Carvajal, andB. Frigo. A simplified method for estimating newmark displacements ofmountain reservoirs.
Soil Dynamics and Earthquake Engineering , 100:518– 528, 2017.[38] L. Wenlian, X. Jianbin, C. Heming, and d. He Tianchun3. Research onthe dynamic response of zhuziqing tailings dam.
Applied Mechanics andMaterials , 170-173:1926–1931, 2012.[39] B. XU, Q. LU, and D. HE. Seismic Stability Analysis of the Yanghuya FlyAsh Tailings Dam.
Environmental and Engineering Geoscience , 20(4):371–391, 11 2014.[40] P. zener, M. Greenfield, S. Sideras, and S. Kramer. Identification of timeof liquefaction triggering.
Soil Dynamics and Earthquake Engineering ,128:105895, 2020.[41] K. Ziotopoulou and R. Boulanger. Calibration and implementation of asand plasticity plane-strain model for earthquake engineering applications.
Soil Dynamics and Earthquake Engineering , 53:268 – 280, 2013.31 ppendix A. Appendix
This appendix includes the seismic records used for all the simulationspresented in this paper, together with the spectrograms calculated to be usedto compute out intensity indicator. All of them has been downloaded from thePEERS seismic database.
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°1
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (a) Record N ◦ Time [Sec] -10-505 A cc e l . [ m / s ] Seismic record N°2
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (b) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°3
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (c) Record N ◦ Time [Sec] -10-505 A cc e l . [ m / s ] Seismic record N°4
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (d) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°5
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (e) Record N ◦ Time [Sec] -50510 A cc e l . [ m / s ] Seismic record N°6
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (f) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°3
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (g) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°8
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (h) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°9
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (i) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°10
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (j) Record N ◦
10 - Northridge-01
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°11
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (k) Record N ◦
11 - Chi-Chi Taiwan
Time [Sec] -10-505 A cc e l . [ m / s ] Seismic record N°12
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (l) Record N ◦
12 - Mammoth Lakes-06
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°13
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (m) Record N ◦
13 - Loma Prieta
Time [Sec] -10-505 A cc e l . [ m / s ] Seismic record N°14
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (n) Record N ◦
14 - Victoria MexicoFigure A.18: Time history acceleration for considered seismic records Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°15
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (a) Record N ◦
15 - Loma Prieta
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°16
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (b) Record N ◦
16 - Chi-Chi Taiwan(TCU071)
Time [Sec] -50510 A cc e l . [ m / s ] Seismic record N°17
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (c) Record N ◦
17 - Coalinga-05
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°18
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (d) Record N ◦
18 - Chuetsu-oki
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°19
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (e) Record N ◦
19 - C. Mendocino
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°20
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (f) Record N ◦
20 - Coalinga-05
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°21
Frequency [Hz] A m p li t ude
10 20 30 40 50 60
Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (g) Record N ◦
21 - Chi-Chi Taiwan(CHY041)
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°22
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (h) Record N ◦
22 - Christchurch NZ
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°23
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (i) Record N ◦
23 - N/A (Ward Fire St)
Time [Sec] -50510 A cc e l . [ m / s ] Seismic record N°24
Frequency [Hz] A m p li t ude
10 20 30 40 50 60 70
Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (j) Record N ◦
24 - N/A (ANGOL)
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°25
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (k) Record N ◦
25 - N/A (PICA)Figure A.19: Time history acceleration for considered seismic records Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°1
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (a) Record N ◦ Time [Sec] -10-505 A cc e l . [ m / s ] Seismic record N°2
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (b) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°3
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (c) Record N ◦ Time [Sec] -10-505 A cc e l . [ m / s ] Seismic record N°4
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (d) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°5
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (e) Record N ◦ Time [Sec] -50510 A cc e l . [ m / s ] Seismic record N°6
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (f) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°3
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (g) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°8
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (h) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°9
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (i) Record N ◦ Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°10
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (j) Record N ◦
10 - Northridge-01
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°11
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (k) Record N ◦
11 - Chi-Chi Taiwan
Time [Sec] -10-505 A cc e l . [ m / s ] Seismic record N°12
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (l) Record N ◦
12 - Mammoth Lakes-06
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°13
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (m) Record N ◦
13 - Loma Prieta
Time [Sec] -10-505 A cc e l . [ m / s ] Seismic record N°14
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (n) Record N ◦
14 - Victoria Mexico
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°15
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (o) Record N ◦
15 - Loma Prieta
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°16
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (p) Record N ◦
16 - Chi-Chi Taiwan(TCU071)
Time [Sec] -50510 A cc e l . [ m / s ] Seismic record N°17
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (q) Record N ◦
17 - Coalinga-05
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°18
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (r) Record N ◦
18 - Chuetsu-okiFigure A.20: Sprectrograms for considered seismic records Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°19
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (a) Record N ◦
19 - C. Mendocino
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°20
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (b) Record N ◦
20 - Coalinga-05
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°21
Frequency [Hz] A m p li t ude
10 20 30 40 50 60
Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (c) Record N ◦
21 - Chi-Chi Taiwan(CHY041)
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°22
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (d) Record N ◦
22 - Christchurch NZ
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°23
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (e) Record N ◦
23 - N/A (Ward Fire St)
Time [Sec] -50510 A cc e l . [ m / s ] Seismic record N°24
Frequency [Hz] A m p li t ude
10 20 30 40 50 60 70
Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (f) Record N ◦
24 - N/A (ANGOL)
Time [Sec] -10-50510 A cc e l . [ m / s ] Seismic record N°25
Frequency [Hz] A m p li t ude Time [s] F r equen cy [ H z ] -15-10-5051015 P o w e r [ d B ] (g) Record N ◦
25 - N/A (PICA)Figure A.21: Sprectrograms for considered seismic records25 - N/A (PICA)Figure A.21: Sprectrograms for considered seismic records