蔡改贫^{a, b}，余 成^a，郝书灏^a，宣律伟^a

（江西理工大学 a. 机电工程学院; b. 江西省矿冶机电工程技术研究中心，江西 赣州 341000）

DOI:10.13732/j.issn.1008-5548.2022.01.003

收稿日期： 2021-09-06，修回日期：2021-09-17，在线出版时间:2021-11-09。

基金项目：国家自然科学基金项目，编号：51464017；江西省重点研发计划项目，编号：20181ACE50034。

第一作者简介：蔡改贫(1964—)，男，教授，博士，博士生导师，研究方向为智能矿山装备技术，智能监控与工业机器人。E-mail：1123615286@qq.com。

摘要：根据不同组分矿物界面的差异性，利用盒子维法实现对矿物界面几何特征的描述、利用原位加载实验实现对矿物界面力学特征的描述，并利用反向传播(back propagation, BP)神经网络算法对矿物界面几何特征和力学特征进行映射预测。结果表明：所研究黑钨矿石中石英-钨界面分形维值大于石英-硅质岩界面的，且石英-钨界面分形维值范围为1.855 1～1.936 8、石英-硅质岩界面分形维值范围为1.163 3～1.371;不同组分粘结界面的力学性质存在差异，且与组成矿物的物理属性、形态特征等相关，其中石英-硅质岩界面最小破碎应力范围为1.178 5～1.482 6 GPa,石英-钨界面最小破碎应力范围为1.335 5～1.542 03 GPa; BP神经网络可实现矿物界面几何特征对力学特征的有效预测，在预测前期误差最大仅为4.14%,随着样本数据增多，预测精度越来越高，在预测后期误差最低仅为0.011%。

关键词：矿物界面；反向传播；神经网络；几何特征；力学特征；破碎特征预测

Abstract：According to the difference of the mineral interface of different components, the box dimension method was used to describe the geometric characteristics of the mineral interface, the in-situ loading experiment was used to describe the mechanical characteristics of the mineral interface, and the back propagation(BP) neural network algorithm was used to describe mapping prediction of the geometric and mechanical characteristics of the mineral interface. The results show that the fractal dimension value of the quartz-tungsten interface in the studied wolframite is larger than the quartz-siliceous interface.The fractal dimension of the quartz-tungsten interface is distributed from 1.855 1 to 1.936 8, and quartz-tungsten's is from 1.163 3 to 1.371. The mechanical properties of the bonding interface of different components are different and related to the physical properties and morphological characteristics of the constituent minerals. Among them, the minimum crushing stress range of the interface of the quartz-siliceous rock is from 1.178 5 to 1.482 6 GPa. The minimum crushing stress range of the quartz-tungsten interface is from 1.335 5 to 1.542 03 GPa. The BP neural network can realize the effective prediction of the geometric characteristics of the mineral interface on the mechanical characteristics, and the maximum error in the early stage of the prediction is only 4.14%. With the increasing of data, the prediction accuracy is getting higher and higher, and the minimum error in the late prediction period is only 0.011%.

Keywords：mineral interface; back propagation; neural network; geometric characteristics; mechanical characteristics; prediction of fracture characteristics

参考文献（References）：

[1]杨松荣, 蒋仲亚, 刘文拯.碎磨工艺及应用[M].北京: 冶金工业出版社，2013.

[2]LI J, WANG F C.Research on the influence of gradation fractal dimension on the compaction characteristics of asphalt mixture based on CT digital image technology[C]//Conference Proceedings of the 8th International Symposium on Project Management, China(ISPM2020).Beijing：China Academic Journal Electronic Publishing House, 2020：372-377.

[3]丁自伟, 李小菲, 唐青豹, 等.砂岩颗粒孔隙分布分形特征与强度相关性研究[J].岩石力学与工程学报，2020，39(9)：1787-1796.

[4]宋宇, 刘保国, 任大瑞, 等.基于分形理论构建随机粗糙节理模型的方法研究[J].岩石力学与工程学报，2021，40(1)：101-112.

[5]LIN M Q，ZHANG L，LIU X Q, et al.The meso-analysis of the rock-burst debris of rock similar material based on SEM[J].Advances in Civil EngineeringVolume, 2020(9168908): 1-13.

[6]XIE J，GAO M Z，ZHANG R, et al.Experimental investigation on the anisotropic fractal characteristics of the rock fracture surface and its application on the fluid flow description[J].Journal of Petroleum Science and Engineering，2020，191(107191)：1-18.

[7]龚爽，赵毅鑫，王震, 等.层理对煤岩动态裂纹扩展分形特征影响试验研究[J].煤炭学报, 2021, 46(8): 2574-2582.

[8]杨阳，李祥龙，杨仁树, 等.低温岩石冲击破碎分形特征与断口形貌分析[J].北京理工大学学报，2020，40(6)：632-639, 682.

[9]刘享华，张科，吴文远.裂隙砂岩破坏过程中的能量耗散与破碎分形特征研究[J/OL].重庆大学学报，2020，1-11[2021-06-05].http://kns.cnki.net/kcms/detail/50.1044.N.20200617.1159.002.html.

[10]刘海峰，朱长歧，汪稔, 等.礁灰岩-混凝土界面剪切特性试验研究[J].岩土力学，2020，41(5)：1540-1548.

[11]AFAQ A，DEMITRIOS M，COTSOVOS N D, et al.Framework for the development of artificial neural networks for predicting the load carrying capacity of RC members[J].SN Applied Sciences, 2020，2(4)：59-68.

[12]LI J F，HU H，LI X, et al.Economic benefit of shale gas exploitation based on back propagation neural network[J].Journal of Intelligent & Fuzzy Systems, 2020，39(6)：8823-8830.

[13]董珍一，林莉，雷明凯, 等.基于BPNN的封严涂层孔隙分布均匀性超声表征[J/OL].航空学报，2021: 1-10[2021-05-14].http://kns.cnki.net/kcms/detail/11.1929.V.20210408.1728.010.html.

[14]邓念东, 丁一, 邢聪聪, 等.基于PCA-BP神经网络的风积砂充填体强度预测[J].矿业研究与开发，2021，41(2)：109-113.

[15]XI X, YIN Z Q.Using artificial neural network to predict the fracture properties of the interfacial transition zone of concrete atthemeso-scale[J].Engineering Fracture Mechanics，2021，242(107488)：1-15.

[16]徐进华.基于灰色系统理论的数据挖掘及其模型研究[D].北京：北京交通大学，2009.

[17]陈健，黄丽华，曲激婷.基于BP神经网络的FRP筋与混凝土界面黏结强度预测[J].大连理工大学学报，2021，61(3)：272-279.

[18]雷廷，贾军元，田福金, 等.基于BP神经网络预测岩石导热系数[J].世界地质，2021，40(1)：131-139.