基于仿真数据的三点弯曲峰值荷载预测方法  

Prediction of Three-point Bending Critical Load Using Artificial Neural Network Based on Simulation Data

作　　者：李立[1,2] 周雷[1,2] 钱紫燕 Li Li;Zhou Lei;Qian Ziyan(State Key Laborulory for the Coul Mine Disaster Dynumics and Conrols,Chongqing Unirersity,Chongqing 400030,P.R.China;School of Resource and Safety Eingineering,Chongqing University,Chonging 400030,P.R.China)

机构地区：[1]重庆大学煤炭灾害动力学与控制国家重点实验室,重庆400030 [2]重庆大学资源与安全学院,重庆400030

出　　处：《地下空间与工程学报》2022年第1期102-111,128,共11页Chinese Journal of Underground Space and Engineering

基　　金：国家自然科学基金(U19B2009)。

摘　　要：三点弯曲试验是广泛采用的抗拉强度参数测试方法之一。通过数值模拟分析发现,岩石三点弯曲峰值荷载与预制裂缝倾角等人为可控因素和杨氏模量、抗拉强度等岩石物性因素都具有正相关关系,而与预制裂缝长度呈现出负相关关系。由于影响因素众多且物理实验工作量大,难以建立多因素数学拟合模型,因此目前尚未有针对峰值荷载的普遍可行的预测方法。基于此,提出在数值模拟的基础上生成大量仿真数据,建立并训练人工神经网络模型,对含预制裂缝岩石三点弯曲试验峰值荷载进行预测。研究表明,基于仿真数据的含预制裂缝岩石三点弯曲峰值荷载神经网络预测模型对于已知数据具有很好的预测性能,并且对于未知数据具有良好的泛化能力和外推能力。Tensile strength of jointed rock mass can be obtained through three-point bending tests of notched specimens. Through numerical simulation, it is posted that critical load of notched rock specimen three-point bending test is subject to artificial parameters including notch angle, and natural parameters including Young’s module, tension strength, which show positive correlations, while there is a negative correlation between critical load and notch length. Limited to numerous factors and large amount of experiment work, it is difficult to build multi-factor mathematical fitting model which can be universally used. This paper set up and trained an artificial neural network model to predict notched rock three-point bending critical load utilizing numerical results of numerical simulation. Simulation data driven artificial neural network model showed reliable prediction performance on known data, and fine ability of generalization and extrapolation on unseen data.

关 键 词：三点弯曲 人工神经网络 仿真数据驱动 参数预测 

分 类 号：TD163[矿业工程—矿山地质测量]