practical diagnosis analysis techniques and behaviour models
TRANSCRIPT
Practical diagnosis analysis techniques and behaviour models
Alexandre Simon (EDF/DTG)Thierry Guilloteau (EDF/DTG)
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A2
Statistical models : HST methodFormulation : Yb = f1(t) + f2(z) + f3(S) + ε (Yb can represent displacement,
piezometric level, flow rate, …)Irreversible trend ( time-dependant) : f1(t) = a1 e-t/t0 + a2 t with t0 : characteristic time of exponential decrease
Hydrostatic reversible influence (depends on hydrostatic load ~z): f2(z) = a3 z + a4 z2 + a5
z3 + a6 z4
Seasonal reversible influence (depends on the season S; S=0 1st January and 2π 31st December) : f3(S) = a7 cos(S) + a8 sin(S) + a9 sin(2 S) + a10 cos(2S)
ε : residuals (inaccuracy of the model)
Reference :Willm G., Beaujoint N. (1967). Les méthodes de surveillance des barrages au service de la production hydrauliqued’Électricité de France. Problèmes anciens et solutions nouvelles. In : Proc. of the Ninth International Congress on Large Dams, Istanbul, 1967, pp. 529-550.
Yb
YcDownstream
Rad
ial D
ispl
ace
me
nt in
mm
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A3
Thermal HST (HSTT)
HST approximation: thermal effect is identical each year.Heat or cold waves are not taken into accountImprovement of HST : integration of air temperature (difference of seasonal temperature and then delayed) Used for displacements of arched dams and thin gravity dams.
FG 582-519
-10,0
-8,0
-6,0
-4,0
-2,0
0,0
1990 1991 1992 1992 1994 1995 1996 1996 1998 1999 2000 2000 2002 2003
Date
Dé
pla
ce
me
nt
(mm
)
aval
Pendule
Analyse HST classique sur 10/01/1990 - 24/12/2003
R = 0,97, S' = 1,22
Reference :Penot I., Daumas B., Fabre J.-P. Electricité De France(2005). Monitoring Behavior. In : Water Power & Dam Construction, December 2005, pp. 24-27.
Yb = f1(t) + f2(z) + f3(S) + k*∆ΘR+ε’
Same expression as HST
Thermal sensitivity of the dam
Differencewithseasonaltemperature, and thendelayed
FG 582-519
-10,0
-8,0
-6,0
-4,0
-2,0
0,0
1990 1991 1992 1992 1994 1995 1996 1996 1998 1999 2000 2000 2002 2003
Date
Dépla
cem
ent (
mm
)
aval
Pendule
Analyse HST Thermique sur 10/01/1990 - 24/12/2003
R = 0,99, S' = 0,60
T90 = 19 jours , Coefficient thermique = -0,53 mm/°C
Témoin température de l'air = MONESTIER
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A4
Statistical models : EFR method
Analysis of piezometric levels which do not react instantaneously with water levelCalculation of a « delayed water level » : takes into account the historic of water levelMethod used for analyze of piezometric level in real timeApplied for zoned-earth dams, rockfill dams
Analyses performed since 1992
upstream dow nstream
Im perv ious ea rth (c lay co re )
Perm eab le earth
Perm eab le ea rth
E qu ipo ten tia l o f po re w a ter p ressure
D ra in
F ilte r
V ib ra ting W ire P iezom ete r
Yb = f1(t) + f3(S) + Ze(T0)+Ze
2 (T0) +Ze3 (T0)+Ze
4 (T0) +ε
Same expression as HST
Delayed hydrostatic effect (T0 : characteristic time)
Yb
HST
EFR
Reference : Fabre J.P., 1992, L’analyse retard, note techique EDF-DTG and CIGB2009, BraziliaAdapatation de l’exploitation des barrages pour maîtriser leur sollicitation : exemples de Laouzas, Gage et Grande Pâture
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A5
The need for new models
New models :HST NL : designed for analysis of piezometric levels affected by fracturation of foundation and non-linear phenomena. Effects (irreversible, hydrostatic, seasonal) are dependant (they are not simply added).
HST HY : designed for analysis of drainage flows and for piezometric levels affected by fluctuation (induced by hydrostatic or thermal effects) of permeability. Effects are also dependant.
HST HY and HST NL can be combined : HST HY NL
Artificial Neural Networks : powerful tool, numerical performance and links between explanatory variables are highlighted (same or better than HST HY NL), thresholds and physical limits of phenomena are taken into account
In developmentUse of air, water or concrete temperature in order to analyze displacements of thin arch dams
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A6
Threshold effect : in summer contact (cracking) between dam and foundation is closed : displacements towards the upstream side are bounded NL = different slopes for evaluation of effect
HST NL (Non Linear)
Linear slope
Infiniteasymptote
Infiniteasymptote
Sigmoid curve
threshold
Differentslopes underthreshold
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A7
HST NL
Decrease of dispersion on corrected measurement : 30% (HST to HST NL)
Hydrostatic effect depends on the season
Seasonal effect depends on the hydrostatic load
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A8
HST HY
Used tu analyse piezometric levels which are affected by cracking of foundation.
Principle : a « length of fracturation » LF is a function of hydrostatic load, season and time :
LF = g1(z) + g2(S) + g3(t) (like HST)
The piezometric level NP depends of LF according to the following equation
NP = Nd+(hydrostatic load(z))*LF where Nd is the piezometric level at downstream
Non linear equation because hydrostatic load depends on z and LF depends on S and t.
Hydrostatic load (resp. seasonal effect or irreversible effects) depends on season (resp. hydrostatic and irreversible effects)
DOWNSTREAMUPSTREAM
Central block
FoundationGrout curtain
Hydrostatic load
Drainage
Upstream decompression cracking
Dowstreamcompression : impervious
Reference : Mauris F., Sausse J., Fabre J.-P., Uplift in arch Dam Foundation, Lessons learned from 42 monitored dams in France , 9th ICOLD European Club Symposium, 10-12 April 2013, Venice, Italy
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A9
HST HY : results
Comparison HST versus HST HYPiezometer 3.3 Roselend
Raw MeasurementsHST correctionHST HY correction
years
Pie
zom
etri
cle
vel
DispersionHST : 4.1 mHST HY : 3,4 m
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A10
Artificial Neural Networks ANN (1/2)
PPG4 (piezometric level) is sensitive to hydrostatic load and the state of cracking of the foundation.
Explanatory variables of ANN : z, cosS, sinS, t; other variables of HST can be easily built by ANN ;
39 parameters (against 10 for HST)
Iterative process to converge (200000 iteration for PPG4)
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A11
Artificial Neural Networks ANN (2/2)Dispersion decrease of 40% (HST versus ANN)
April : dam is at its coldest foundation, cracking reaches its maximum, higher slope for hydrostatic load (red)
October : maximum thermal expansion of the dam, state of cracking at its minimum, slope of hydrostatic effect is the slowest
Same effect can be observed on the seasonal effect
HST-corrected data still shows a seasonal aspect correction. This cyclical aspect has disappeared with ANN
Best performance of ANN thanks to the seasonally-dependant hydrostatic effect.
Reference :Simon A., Royer M., Mauris F., Fabre J.P. , Electricité de France, Analysis and Interpretation of Dam Measurements using Artificial Neural Networks, 9th ICOLD European Club Symposium, 10-12 April 2013, Venice, Italy
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A12
Multi-component analyze : PCA Principal Component Analysis
Example of use : displacement measured by pendulums (corrected measurements) ; aim of PCA : reducing number of sensors and identifying links between sensors
Division Production Ingénierie Hydraulique - DTG Monday, May 12, 2014H-44200965-2014-000190-A13
Conclusion
1967 1992 2004 2008-2012
HST EFR HSTT
HST HYHST NLANN
Raw Data are still analyzed (Sometimes HST can not be performed)Thanks to the development of micro-computer EFR was developed in the early 90’sHeat wave in 2003 : development of Thermal HSTAnalysis of state of cracking for arch dams : HST HY NL and ANN