论文部分内容阅读
将颗粒流程序应用于解决工程问题的前提是对岩石试件的几个力学参数进行准确标定,抗拉强度的标定便是其中之一。通过比较分析巴西实验和直接拉伸实验,结合具有非线性映射能力的BP神经网络,对颗粒流岩石的细观参数进行反演研究。表明:巴西实验条件苛刻,难以保证大量实验成功进行,无法向神经网络提供高质量训练样本,反演精度仅为61%;直接拉伸实验条件宽松,可以保证神经网络训练样本的数量和质量,反演精度可提高到83%;在保证样本数量和质量的前提下,BP神经网络有能力实现宏细观参数的准确映射,是颗粒流岩石参数标定的有效手段。
The particle flow program is applied to solve the engineering problem on the premise that several mechanical parameters of the rock specimen are accurately calibrated, and the calibration of tensile strength is one of them. By comparing and analyzing the Brazilian experiment and the direct tensile experiment, the BP neural network with nonlinear mapping ability is used to invert the mesoscopic parameters of the particle flow rock. The experimental results show that the experimental conditions in Brazil are harsh and it is difficult to ensure that a large number of experiments can be successfully carried out and the high-quality training samples can not be provided to the neural network. The inversion accuracy is only 61%. The experimental conditions of direct tension are loose and can guarantee the quantity and quality of neural network training samples. The accuracy of inversion can be improved to 83%. The BP neural network has the ability to accurately map the macroscopic parameters under the premise of ensuring the quantity and quality of the samples. It is an effective means to calibrate the particle flow parameters.