The parameters of the constitutive model, the creep model, and the wetting model of materials of the Nuozhadu high earth-rockfill dam were back-analyzed together based on field monitoring displacement data by employing an intelligent back-analysis method. In this method, an artificial neural network is used as a substitute for time-consuming finite element analysis, and an evolutionary algorithm is applied for both network training and parameter optimization. To avoid simultaneous back-analysis of many parameters, the model parameters of the three main dam materials are decoupled and back-analyzed separately in a particular order. Displacement back-analyses were performed at different stages of the construction period, with and without considering the creep and wetting deformations. Good agreement between the numerical results and the monitoring data was obtained for most observation points, which implies that the back-analysis method and decoupling method are effective for solving complex problems with multiple models and parameters. The comparison of calculation results based on different sets of back-analyzed model parameters indicates the necessity of taking the effects of creep and wetting into consideration in the numerical analyses of high earth-rockfill dams. With the resulting model parameters, the stress and deformation distributions at completion are predicted and analyzed.
A large number of high earth-rockfill dams located in western China with heights of 250 m to 300 m are currently under construction or being planned. Among these dams, the Nuozhadu earth-rockfill dam, with a height of 261.5 m, is the highest earth-core rockfill dam under construction in China. To ensure safety, a large number of observation instruments have been installed at different elevations and different zones in the dam during the construction period. To date, field observation data have been collected to analyze the characteristics of dam materials and the stress-deformation distribution in the dam and to facilitate the prediction of future deformation.
The calculation of earth-rockfill dam deformation is affected by many factors, such as the representativeness of soil samples, the size effect of laboratory tests, the differences of sample preparation and loading conditions from the real construction conditions, and the imperfection of the constitutive model and the numerical method. Moreover, along with the construction process, the model parameters of dam materials will change with time due to the breakage and wetting of rockfill particles. Therefore, it is of great importance to dynamically back-analyze the model parameters of dam materials based on field observation data to improve the accuracy of deformation prediction.
Displacement back-analysis is an effective method to identify the model parameters of soils and rocks. In conventional back-analysis methods, the optimal values of parameters are usually progressively approximated by minimizing the error function through iterations. In general, the range and initial values of the parameters should be given before the analysis, the time-consuming finite element method (FEM) calculation is performed frequently, the rate of convergence is slow, and sometimes the back-analysis fails for large-scale nonlinear problems. Furthermore, the result is often affected by the initial values and a local minimum or premature convergence is likely to be obtained. Therefore, for large-scale multiparameter nonlinear problems, the solution is sometimes unstable. In recent years, the artificial intelligence back-analysis method was introduced to geotechnical engineering. With the development of intelligent optimization algorithms, the artificial intelligence back-analysis method is continuously further improved. Extensive studies have been conducted to develop different displacement-based back-analysis methods [
To date, studies of displacement-based back-analysis methods have mainly focused on underground engineering and rock mass, whereas studies on dam projects are relatively scarce. In addition, there have been few studies on the back-analyses of dams in the process of construction, which are usually performed after the construction is completed. In addition, the displacement-based back-analyses, usually focused on the constitutive model parameters, pay little attention to the parameters of wetting and creep models. In particular, currently, there are no research results concerning the back-analysis of wetting deformation in geotechnical engineering.
Wetting deformation and creep deformation, for which many numerical calculation models and methods have been built, have great significance in the stress redistribution and stability of earth-rockfill dams. The mechanism of wetting deformation is generally investigated using laboratory tests [
Nuozhadu dam is the first high earth-rockfill dam with comprehensive monitoring. The collected observation data of Nuozhadu dam fully reflect the state of the dam and are of great significance for the study of the stress and deformation characteristics in high earth-rockfill dams. In this study, an intelligent back-analysis method based on artificial neural networks and evolutionary algorithm [
The Nuozhadu hydropower station is located on the main stream of the lower Lancang River, near Pu’er City of Yunnan Province. The total installed capacity is 5,850 MW, and the designed annual average power output is 239 × 108 kW
The maximum cross-section of the dam, with material zoning, is shown in Figure
The maximum cross-section of Nuozhadu earth-core rockfill dam.
The construction process of the core and the impounding process of the reservoir are shown in Figure
Construction process of the dam and water level of the reservoir.
Observation instruments were installed on several cross-sections of the dam. The layout of the observation instruments on the maximum cross-section is shown in Figure
The stress-strain relationship, creep behavior, and wetting deformation of the dam materials are needed to simulate the behavior of the dam during the construction and impounding process. Duncan and Chang’s E-B model [
This model was developed to provide a simple framework encompassing the most important characteristics of soil stress-strain behavior. The nonlinear stress-strain curves are represented by hyperbolae, whose instantaneous slope is the tangent modulus,
The bulk modulus can be expressed as
The Mohr-Coulomb envelopes for almost all soils are curved to some extent, and the wider the range of pressure involved the greater the curvature, especially for cohesionless soils such as sand, gravel, and rockfill. For example, in the bottom near the center of a large dam, rockfill may be confined under such a large pressure that the friction angle may be several degrees smaller than that near the surface of the slopes. This variation in property may be represented by an equation of the form
It can be seen that there are seven parameters in Duncan and Chang’s E-B model, that is,
The seven-parameter creep model is commonly used in the numerical analyses of earth dams. Merchant’s equation is used to describe the rheological deformation curve in the creep model:
Overall, there are seven parameters in the creep model [(
A modified Shen’s three-parameter wetting model was used to calculate the wetting deformation of the dam materials. The wetting deformation in the model consists of two components, the volumetric wetting deformation
Because of the complexity of geotechnical engineering problems, conventional parameter back-analysis methods often require a large number of forward finite element analyses and thus a long computation time, and the result may be easily trapped in local minimum values. In the back-analysis method [
Figure
Displacement back-analysis method based on neural network and evolutionary algorithm.
The displacement back-analysis software EBA-EANN, developed based on this method, has been successfully applied to several earth-rockfill dams in China with good results [
An FEM model was used to calculate the stress and deformation response of the dam and to generate the training samples for the neural network. Figure
3D FEM mesh of Nuozhadu earth-core rockfill dam.
Displacement back-analyses were performed at two different construction stages (see Figure
Stepwise displacement back-analyses.
Back-analysis | Observation data date | Upstream water level | Parameters analyzed |
---|---|---|---|
1 | Dec. 15, 2011 | 666 m |
|
|
|||
2 | May 14, 2012 | 738.3 m |
|
Measured displacements at typical observation points.
With sensitivity analysis of the model parameters, only the four main E-B model parameters, that is,
Main E-B model parameters of main dam materials.
Material |
|
|
|
|
---|---|---|---|---|
Rockfill I | ||||
Test | 1425 | 540 | 0.26 | 0.16 |
Back 1 | 1246 | 411 | 0.14 | 0.11 |
Rockfill II | ||||
Test | 1400 | 620 | 0.17 | 0.05 |
Back 1 | 1188 | 393 | 0.145 | 0.043 |
Gravelly clay | ||||
Test | 320 | 210 | 0.48 | 0.26 |
Back 1 | 368 | 244 | 0.226 | 0.038 |
To avoid this problem, the back-analyses of different dam materials were decoupled by considering material zoning, construction process, and observation point locations. First, most of the water level settlement and wire alignment transducer observation points at EL. 626 m are in the downstream rockfill I zone. As this region is at the bottom of the dam and was constructed at an earlier time, the displacement distribution in this region mainly depends on the model parameters of rockfill I, whereas other materials with given density act as loading on this region. Therefore, the model parameters of rockfill I could be back-analyzed separately from displacement measurements in the rockfill I zone at EL. 626 m. Then, with the obtained model parameters of rockfill I, the model parameters of rockfill II could be back-analyzed from displacement measurements at EL. 660 m and EL. 701 m, except for the measurements in the core wall. Finally, the model parameters of gravelly clay could be back-analyzed using the settlement measurements of electromagnetic gauges in the core wall. With this treatment, the number of samples was reduced to 34 × 3 = 243, which is much lower than the simultaneous back-analysis number.
Reasonable observation data of selected measurement points were used as the targets of back-analysis. The measurement points were selected on the basis of previous numerical analyses, quality of observation data, and experiences of numerical calculation. The locations of the measurement points used in the 1st back-analysis are shown in Figure
The measurement points used in the 1st back-analysis.
The results of the 1st back-analysis are also listed in Table
Comparison between calculated results (1st back-analysis parameters) and observation data.
To investigate the effects of creep and wetting deformation, the creep and wetting model parameters, as well as the two main E-B model parameters (
The E-B and creep model parameters of the three main dam materials were decoupled as before. The locations of the measurement points used in the 2nd back-analysis are shown in Figure
Main E-B and creep model parameters.
Material |
|
|
|
|
|
---|---|---|---|---|---|
Rockfill I | |||||
Test | 1425 | 540 | 0.00600 | 1.00 | 0.00423 |
Back 2 | 1486 | 665 | 0.00314 | 1.61 | 0.00311 |
Rockfill II | |||||
Test | 1400 | 620 | 0.00600 | 1.00 | 0.00612 |
Back 2 | 1643 | 717 | 0.00300 | 2.17 | 0.00821 |
Gravelly clay | |||||
Test | 320 | 210 | 0.00300 | 1.00 | 0.00717 |
Back 2 | 510 | 340 | 0.00345 | 1.27 | 0.00849 |
The measurement points used in the 2nd back-analysis.
The wetting model parameters of rockfill I and rockfill II were back-analyzed together, and the measurement points used in the back-analysis of wetting model parameters are DB-C-VW-10, DB-C-VW-11, DB-C-VW-12, and DB-C-VW-13 (see Figure
Wetting model parameters.
Parameters | Rockfill I | Rockfill II | ||||
---|---|---|---|---|---|---|
|
|
|
|
|
| |
Test | 2.820 | 1.730 | 0.362 | 2.980 | 1.780 | 0.356 |
Back 2 | 1.417 | 0.869 | 0.904 | 1.493 | 0.892 | 0.890 |
Through field examination, it was found that the compaction degree of the gravelly clay is generally better than the designed value, which, to a certain degree, justifies the high deformation moduli obtained by the back-analyses. In Figure
Comparison of displacement values (mm).
Observation point | Observed | Calculated | |
---|---|---|---|
Back | Test | ||
DB-C-V-04 | 217.0 | 222.2 | 192.8 |
DB-C-VW-03 | 390.0 | 393.2 | 335.2 |
DB-C-H-05 | 198.9 | 203.4 | 168.6 |
DB-C-SR-31 | 664.2 | 698.6 | 830.2 |
Comparison between calculated results (2nd back-analysis parameters) and observation data.
In the stepwise back-analyses of Nuozhadu dam, the creep and wetting deformations were not considered in the first analysis. Although the calculation results based on the first back-analyzed parameters agree well with the observation data before impounding, the trends of calculated displacements can be different from that of the observation data in the later period. To illustrate the influence of creep and wetting deformations, Figure
Comparison of calculation results based on the two sets of back-calculated parameters through construction completion.
With the back-calculated model parameters, the displacement and stress distributions at completion were predicted (Figure
Displacement and stress distribution at completion.
The deformation observation data of the Nuozhadu high earth-rockfill dam, which fully reflects the state of the dam, plays an important role in analyzing the characteristics of the dam materials and facilitating the prediction of future deformation. In this study, the model parameters of Duncan and Chang’s E-B model, the seven-parameter creep model, and a modified Shen’s three-parameter wetting model of the Nuozhadu high earth-rockfill dam were back-analyzed based on field monitoring displacement data by employing an intelligent back-analysis method. Two displacement back-analyses have been performed at different construction stages, with and without considering the creep and wetting deformations. To avoid simultaneous back-analysis of many parameters, the model parameters of the three main dam materials are decoupled and back-calculated separately according to material zoning, construction process, and observation point locations. The resulting numerical data are in good agreement with the monitoring data for most observation points. The deviation of the model parameters from the laboratory tests revealed by the stepwise back-analyses has been partially verified by field examination results. The back-analysis method and decoupling method used in the back-analysis were effective at addressing complex problems with multiple models and parameters. The comparison of calculation results based on different sets of back-calculated parameters indicates that the breakage of particles and impounding will cause certain deformation, and it is necessary to take the effects of creep and wetting into consideration in the numerical analyses of high earth-rockfill dams. With the back-calculated parameters, the stress and deformation distributions at completion were predicted and analyzed, from which conclusive results were obtained.
The authors declare that there is no conflict of interests regarding the publication of this paper.
The authors would like to thank the financial support of the National Natural Science Foundation of China nos. 51209118, 51179092, and 51379103 and State Key Laboratory of Hydroscience and Engineering Project no. 2013-KY-4.