针对带式输送机巡检机器人在煤矿井下的环境适应性、自主快速安全充电、基于音视频分析的带式输送机异常工况智能识别等关键技术问题，阐述了煤矿带式输送机巡检机器人系统总体设计；设计了一种模块化、两节车厢挂载式巡检机器人本体及吊挂单轨式、滚轮与链轮链条混合驱动的巡检机器人行走机构，可适应煤矿井下输送带运输复杂环境；提出了煤矿井下电—机—电能量转换充电方法，设计了基于鼓形齿的充电自动对接机构，可实现巡检机器人在煤矿井下自主快速安全充电，满足井下远距离巡检需求；提出了基于托辊运转的声压1/3倍频程谱及声品质尖锐度特征指标的托辊故障识别方法，针对井下多粉尘、浓雾气、低照度场景下视频质量差导致视频识别难的问题，提出了自适应图像增强算法框架，增强图像质量，在此基础上，基于目标轮廓先验知识，构建深度神经网络模型，实现煤矿井下人员、煤流、带面异物、带面损伤等特征不显著目标的视频智能识别，提高带式输送机巡检机器人智能化水平。对带式输送机巡检机器人样机及带式输送机异常工况音视频样本进行了验证测试，结果表明：巡检机器人最小转弯半径1 m，在负重220 kg、巡检速度0.4 m/s的情况下，可在角度为20°的坡道上平稳运行；在充电站端的电动机额定功率2.2 kW情况下，机器人本体端的发电机能稳定输出功率800 W；基于音视频分析的带式输送机异常工况智能识别方法具有应用可行性。煤矿带式输送机巡检机器人关键技术的突破，将加速带式输送机巡检机器人的推广应用，有力促进智能化矿井建设。

In view of the key technical problems such as the adaptability of the inspection robot of the belt conveyor in the coal mine,the independent fast and safe charging,the intelligent recognition of the abnormal working condition of the belt conveyor based on the analysis of audio and video,this paper expounds the overall design of the inspection robot system of the belt conveyor in the coal mine,and designs a modular,two compartment mounted inspection robot body and a single track,roller and chain wheel hybrid drive inspection robot walking mechanism,which can adapt to the complex environment of underground belt transportation; the method of underground electric machine electric energy conversion charging is proposed,and the automatic charging docking mechanism based on drum shaped teeth is designed,which can realize the inspection robot′s independent fast and safe charging in coal mine and meet the requirements of long-distance inspection in coal mine; Based on the 1 / 3 octave frequency spectrum of sound pressure and the characteristic index of sound quality sharpness,a fault identification method of roller is proposed,In view of the problem that video quality is poor in the scene of multi dust,dense fog and low illuminance,which makes video recognition difficult,an adaptive image enhancement algorithm framework is proposed to enhance image quality. On this basis,based on the prior knowledge of target contour,a depth neural network model is constructed,to realize the video intelligent recognition of the characteristics of coal mine underground personnel,coal flow,foreign matters on the belt surface,damage on the belt surface and so on,so as to improve the intelligent level of inspection robot of belt conveyor. The prototype of inspection robot for belt conveyor and the audio and video samples of abnormal working conditions of belt conveyor are tested. The results show that:the minimum turning radius of inspection robot is 1 m,under the condition of 220 kg load and 0.4 m/s patrol inspection speed,it can run smoothly on the ramp with an angle of 20°,under the condition that the rated power of the motor at the charging station is 2.2 kW,the generator at the robot body end can stably output power of 800 W. the intelligent recognition method of belt conveyor′s abnormal working condition based on audio-video analysis is feasible. The breakthrough of the key technology of the inspection robot of the belt conveyor in the coal mine will accelerate the popularization and application of the inspection robot of the belt conveyor and effectively promote the construction of the intelligent mine.