边坡稳定性强度折减颗粒离散元法分析的细观参数标定策略

Calibration of Micro Parameters of Particles in Granular Discrete Element Method to Assess Slope Stability by Strength Reduction Method

根据岩土体力学指标标定颗粒细观参数是应用颗粒离散元解决岩土工程问题的一项基础工作。利用颗粒离散元法执行边坡稳定性强度折减法分析时,岩土体抗剪强度随折减系数变化不断调整,使用试算法标定颗粒细观参数效率严重不足。为解决此问题,采用国产颗粒离散元软件MatDEM,以颗粒细观参数为输入,以岩土体抗剪强度指标为输出,构建超定BP神经网络,开发双轴压缩数值模型并行试验技术,加速神经网络样本数据的获取过程,重复执行“逆向标定—精度检查—样本修正”以实现颗粒细观参数的逆向迭代修正标定。结果表明:该策略的精度明显优于直接以抗剪强度指标为输入、以颗粒细观参数为输出的欠定BP神经网络,标定的颗粒细观参数与抗剪强度目标高度匹配,数值试验结果与目标值相比误差可控制在不超过1%的范围内。采用澳大利亚计算机应用协会(ACADS)的均质土质边坡和堆石坝等两个边坡稳定性分析经典考题,以模型平均位移突变为极限状态判据执行强度折减法分析,检验该策略的应用效果。结果显示:该策略的标定能力可满足抗剪强度不断调整时颗粒细观参数重新设定的需要,安全系数计算结果与ACADS推荐解具有良好可比性。该策略可为MatDEM及其他颗粒离散元程序执行边坡稳定性强度折减法分析时颗粒细观参数标定提供可靠途径。

Calibration of micro parameters of particles based on the mechanical properties of rock and soil mass is essential to solve various geo-technical problems by using the granular discrete element method.If the strength reduction method is employed in the granular discrete element method to assess slope stability,the shear strength of rock and soil mass will be continuously adjusted according to the reduction factor,and then,micro parameters of particles should be constantly calibrated.In this case,the trial-and-error method is cumbersome and time-consuming.To solve the issue,take MatDEM as an example,an overdetermined BP neural network was built by setting micro parameters of particles as input and the shear strength of rock and soil mass as output.A technique was developed to carry out biaxial compression numerical tests in parallel.Then,a reverse-iterative-correct strategy was proposed to calibrate micro parameters of particles for a prescribed value of shear strength,by executing the“reversely determination—error check—sample correction”process repeatedly.Results of numerical tests verified that the new strategy apparently had a higher precision than the underdetermined BP neural network by taking the shear strength as input and micro parameters as output.Micro parameters obtained by the new strategy had a remarkable level of compliance with the specified shear strength,and the relative error between the prescribed values and the numerical test results was less than one percent.Two exam problems,a homogeneous soil slope and a rock-fill dam,suggested by ACADS are adopted to verify the ability of the proposed strategy in slope stability analysis based on the strength reduction method.Results showed that the new strategy satisfied the calibration requirement on micro parameters of particles when executing the strength reduction method in MatDEM and the resulting safe factors agree with the safe factors recommended by ACADS.The proposed strategy provides a reliable approach for MatDEM and other granular discrete element software to calibrate micro parameters when assessing slope stability by the strength reduction method.