摘要
虽然增强学习方法能够为井下铲运机自主铲装作业提供解决方案,但直接将其应用在真实环境中进行数据采集存在诸多问题。为满足自主铲装作业训练对数据的需求,借鉴某型号正转四杆结构的井下铲运机,在V-REP中建立铲运机、矿石爆堆及其工作环境的动力学模型,并通过编写控制脚本实现外接设备操纵铲运机作业,仿真和模拟矿石爆堆的生成,从而构建了仿真试验平台。结果证明在虚拟环境下可以实现对井下铲运机在溜井处铲装矿石,并经由运输巷道运输至溜井处卸载矿石这一完整作业流程的仿真。虚拟环境中模型的运动状态良好、模型间交互正常,可作为后续增强学习框架下自主铲装作业的训练环境,也可作为井下作业人员上岗前的培训平台。
In order to avoid many unfavorable factors in the traditional environment of mining operations, improve production efficiency of mines,reduce the labor cost and eliminate the accident potentials,the large mining equipment such as underground LHD(load-haul-and-dump-machine) is developing towards the direction of intelligence,information and unmanned. The autonomous scooping operation of LHD is an important part of relative researches,but the scooping operation is difficult to predict due to the complex structure of the working device of LHD.With the rapid development of methods such as reinforcement learning in recent years,there are some feasible solutions to underground LHD’s autonomous scooping operation. However,a large amount of training data are required to be collected due to the limits of relative algorithms and there are several issues such as long acquisition time,high cost of existing equipment transformation and interference in normal operation of mines when collecting data in real environment.Under the existing technical conditions,virtual prototype technology can be used to build a simulation operation platform to meet the demand of data for autonomous scooping operation training. In this paper,the structure of a certain type of underground LHD with forward rotation four rods mechanism was taken as reference to analyze the linkages of working devices and articulated steering mechanism of the car body and a three-dimension dynamic model of LHD was established in V-REP combined with the map model generated by 3D Studio Max.12 basic ore models with 3 irregular shapes in 4 sizes were established to form blasted-pile randomly under control of scripts.Then, generated an interactive working environment including tunnel,drawing points,ore pass and simulated light sources to establish an simulation experiment platform combined with LHD model and blasted-pile and wrote embedded script to realize to communication between the external equipment and the simulation experiment platform to control walk and scooping operation of LHD and generation of blasted-pile.Finally,finished the real- time simulation of underground LHD’s scooping and loading ores at drawing point and transported it to the ore pass through the transportation tunnel.The simulation results show that the loading and unloading operations can be simulated under the virtual environment after practical test to meet the demands of data under reinforcement learning framework and the simulation platform can also be used as a training platform for underground workers.
作者
吴家希
王李管
李亚龙
WU Jiaxi;WANG Liguan;LI Yalong(School of Resources and Safety Engineering,Central South University,Changsha 410083,Hunan,China;Digital Mine Research Center,Central South University,Changsha 410083,Hunan,China)
出处
《黄金科学技术》
CSCD
2020年第1期124-133,共10页
Gold Science and Technology
基金
国家重点研发计划项目“深部集约化开采生产过程智能管控技术”(编号：2017YFC0602905)资助
