Multi-parameter inverse analysis of concrete dams using kernel extreme learning machines-based response surface model

Grain morphology has significant impacts on the mechanical behaviors of granular materials. However, its influences on grain breakage are still poorly understood due to the complexity of morphological characterization. This work employs the machine learning (ML) approach to reveal the impacts of grain morphology on grain crushing strength. A database consisting of 400 high-resolution rock grains was collected by 3D scanning. The morphology of each grain is characterized using 24 global shape descriptors and the local roundness at the contact zone. Based on hierarchical clustering, eight descriptors are selected as input features to the ML model. Grain instability is also quantified and used in the training of the ML model. To eliminate the influence of material properties and size effects on grain breakage, we perform the combined finite and discrete element method simulation of the single grain crushing test to get the crushing strength, which is used as the output of the ML model. Among the five ML regression algorithms used, XGBoost has more outstanding prediction accuracy and generalization ability. Each feature's influence on crushing strength is then analyzed using permutation feature importance and SHapley Additive exPlanations (SHAP) value. The feature importance ranking indicates that grain instability has the most significant impact on the crushing strength. The SHAP value demonstrates that flat, rounded, and convex grains have higher resistance to compression. The local roundness at the contact points also affects the crushing strength, in which higher local roundness corresponds to larger crushing strength. The results revealed by ML are in agreement with previous qualitative studies. The interpretable ML may shed new light on the investigation of the complex granular materials.

Machine learning reveals the influences of grain morphology on grain crushing strength

High concrete-face rockfill dams (CFRDs) with heights of over 100 m have been quickly developed in recent years. The self-weight of rockfill materials causes creep deformation of the dam body. However, the creep analysis method of high CFRDs in finite element software is few, and sometimes, it can also not reflect the long-term performance of high CFRDs well. Therefore, it is necessary to carry out the secondary development in finite element software. This study developed a subroutine that can run in Finite Element Method (FEM) platform ABAQUS to simulate long-term creep deformation behavior of the rockfill materials more accurately. Then, a displacement back-analysis for parameters, based on the Xujixia high CFRD project, is performed by the neural network response surface method (BP-MPGA/MPGA). Remarkable agreements are observed between simulation and field monitoring results. The calibrated FEM model is used to predict stress and deformation behavior of the Xujixia high CFRD after three years of operation period. The result indicates that rockfill creep deformation has a significant impact on stress and deformation of the high CFRD during the operation. This research may predict long-term performance using FEM in the design stage for high CFRDs.

/pdf/long-term-settlement-of-high-concrete-face-rockfill-dam-by-vzfcedesr8.pdf

Long-Term Settlement of High Concrete-Face Rockfill Dam by Field Monitoring and Numerical Simulation

The mechanical parameters of materials in a dam body and dam foundation tend to change when dams are reinforced in aging processes. It is important to use an early-warning index to reflect the safety status of dams, particularly of heightened projects in the impoundment period. Herein, a new method for monitoring the safety status of heightened dams is proposed based on the deformation monitoring data of a dam structure, a statistical model, and finite-element numerical simulation. First, a fast optimization inversion method for estimation of dam mechanical parameters was developed, which used the water pressure component extracted from a statistical model, an improved inversion objective function, and a genetic optimization iterative algorithm. Then, a finite element model of a heightened concrete gravity dam was established, and the deformation behavior of the dam with rising water levels in the impoundment period was simulated. Subsequently, mechanical parameters of aged dam parts were calculated using the fast optimization inversion method with simulated deformation and the water pressure deformation component obtained by the statistical model under the same conditions of water pressure change. Finally, a new early-warning index of dam deformation was constructed by means of the forward-simulated deformation and other components of the statistical model. The early-warning index is useful for forecasting dam deformation under different water levels, especially high water levels.

Deformation early-warning index for heightened gravity dam during impoundment period

To evaluate the real working behaviour of high arch dams, the in-situ monitoring data-based inversion of deformation parameters is essential. The frequently used hydraulic, seasonal, and time model cannot accurately interpret the viscoelastic deformation behaviour of concrete dams. The main purpose of this paper is to propose a hydraulic, exponential, seasonal, and time (HEST) statistical model to conduct the multi-point zonal intelligent inversion of viscoelastic parameters of high arch dams. The extreme learning machine (ELM) is used to establish the nonlinear relationship between viscoelastic parameters and finite element method-calculated total hydraulic displacement. The particle swarm optimization is improved by three strategies and is then used to search the optimal viscoelastic parameters from the ELM model, according to the HEST model-separated total hydraulic displacement. An engineering example of the Jinping-I arch dam has demonstrated the good performance of the proposed HEST model and inversion method. In the water level stable stage, the abnormal increase of dam radial displacement, which is measured by the plumb line monitoring system, can be reasonably interpreted by the air temperature annual drop effect and dam concrete's hysteretic deformation, and the latter can be reasonably represented by the step-type exponential function-based hysteretic hydraulic component, which occupies a proportion of 20–30%. The inverted instantaneous elastic moduli of concrete in the four clustered zones of the dam body are 44.8, 48.8, 30.8 and 41.0 GPa, respectively, and other viscoelastic parameters are linearly proportional to these. The inverted parameters in the central of the dam body are the highest, followed by the crest region, while smaller values are obtained for dam regions near the foundation. With the utilization of deformation spatial association, the zonal intelligent inversion can effectively avoid the interference of the location of used dam displacement monitoring points.

Zonal intelligent inversion of viscoelastic parameters of high arch dams using an HEST statistical model

Due to the size effects of rockfill materials, the settlement difference between numerical simulation and in situ monitoring of rockfill dams is a topic of general concern. The constitutive model parameters obtained from laboratory triaxial tests often underestimate the deformation of high rockfill dams. Therefore, constitutive model parameters obtained by back analysis were used to calculate and predict the long-term deformation of rockfill dams. Instead of using artificial neural networks (ANNs), the response surface method (RSM) was employed to replace the finite element simulation used in the optimization iteration. Only 27 training samples were required for RSM, improving computational efficiency compared with ANN, which required 300 training samples. RSM can be used to describe the relationship between the constitutive model parameters and dam settlements. The inversion results of the Shuibuya concrete face rockfill dam (CFRD) show that the calculated settlements agree with the measured data, indicating the accuracy and efficiency of RSM.

Long-term deformation analysis of Shuibuya concrete face rockfill dam based on response surface method and improved genetic algorithm

A predictive deep learning framework for path-dependent mechanical behavior of granular materials

A set of numerical experiments of simple shear tests of rockfill–structure interfaces were performed using the combined finite- and discrete-element method (FDEM). Each particle was modeled...

Shao-heng Guan

Papers

Long-term deformation analysis of Shuibuya concrete face rockfill dam based on response surface method and improved genetic algorithm

A predictive deep learning framework for path-dependent mechanical behavior of granular materials

Interfacial Shearing Behavior Analysis of Rockfill Using FDEM Simulation with Irregularly Shaped Particles

A machine learning-based multi-scale computational framework for granular materials

A predictive deep learning framework for path-dependent mechanical behavior of granular materials