矿山运输中，带式输送机是煤炭工业广泛使用的运输设备，带式输送机在运行过程会出现各种各样的异常，给煤矿生产和工作人员安全带来隐患。托辊作为带式输送机数量最多的部件，由托辊异常所引起的带式输送机异常占总异常的1/3左右。本文通过对托辊异常种类和成因机理深入分析，针对已有异常检测过程中存在的问题，设计了托辊异常自动检测方法，主要工作如下。

针对已有托辊检测方法中存在的问题，提出基于视频数据的托辊异常检测方法。基于声音的检测方法，卡死托辊无法有效检测到故障信号，漏检严重；基于压力、温度和电压等需要大量传感器进行数据采集的检测方法，传感器的安装和维护成本太高。基于视频数据检测方法主要依据托辊速度反映了托辊运动状态，使用巡检机器人拍摄托辊视频，从视频中估算出托辊旋转线速度，比较估算出的托辊速度和皮带速度实现托辊异常检测，该算法的核心是从采集的视频中估算托辊旋转线速度。

针对使用线速度公式计算托辊速度时托辊旋转频率不易直接测量的问题，提出使用快速傅叶变换的时频分析方法计算托辊旋转频率。首先通过提取连续视频帧中托辊灰度纹理特征将托辊周期旋转变化过程转化为一系列随时间变化的周期性数字序列。然后使用快速傅里叶变换FFT将数字序列由时域变换到频域得到对应频谱图，根据频谱图确定托辊旋转频率。

针对分步计算托辊速度时、由于误差累积导致的最终托辊速度计算不准确问题，提出使用3D卷积网络端到端对托辊速度进行检测，根据托辊速度和皮带速度比值判断托辊状态，实现托辊的异常检测。实验中使用3D卷积网络提取视频中托辊的时空特征，根据提取特征结果对待测托辊速度进行分类。创建用于深度学习的托辊数据集，采用多种数据集扩充方法将原始数据集扩充为原来的10倍大小，提高了模型的检测精度。最后还使用了五个评价指标对3D托辊速度分类模型进行了评价，模型评价结果满足实验预期。

基于视频的托辊异常检测实现了托辊异常的自动化检测，降低了巡检工人的负担，及时发现异常托辊。相比于需要大量传感器的检测方法和基于声音信号特征的托辊异常检测方法，基于视频的检测方法降低了传感器安装和维护难度，提高卡阻托辊检测准确率，确定异常托辊的位置，降低工人维修时二次定位异常托辊的难度和工作量。

In mine transportation, belt conveyors are widely used transportation equipment in the coal industry. Various abnormalities may occur during operation of belt conveyors, which brings hidden dangers to coal mine production and worker safety. As the belt conveyor with the largest number of components, the idler rollers account for about 1/3 of the total abnormalities caused by the abnormality of the idler rollers. In this paper, through in-depth analysis of the types and causes of the abnormality of the roller, in view of the existing problems in the abnormal detection scheme, an automatic detection method for the abnormality of the roller is designed. The main work is as follows.

Aiming at the problems existing in the existing idler detection methods, an abnormal detection method for idler rollers based on video data is proposed. The sound-based roller abnormal detection method, the stuck roller cannot effectively detect the fault signal, and the missed detection is serious; based on the pressure, temperature, and voltage detection methods that require a large number of sensors for data collection, the installation and maintenance costs of the sensors are too high. The detection method based on video data mainly reflects the movement state of the idler according to the speed of the idler. The inspection robot is used to shoot the idler video, and the linear speed of the idler is estimated from the video, and the estimated idler speed is compared with the belt speed to realize the support Roller abnormality detection, the core of this algorithm is to estimate the linear speed of roller rotation from the collected video.

Aiming at the problem that it is not easy to directly measure the rotation frequency of the idler when using the linear velocity formula to calculate the idler speed, a Fast Fourier Transform (Fast Fourier Transform) time-frequency analysis method is proposed to calculate the rotation frequency of the idler. First, by extracting the gray texture features of the roller in the continuous video frames, the roller's periodic rotation change process is transformed into a series of periodic digital sequences that change with time. Then, use the fast Fourier transform to transform the digital sequence from the time domain to the frequency domain to obtain the corresponding spectrogram, and determine the rotation frequency of the roller according to the spectrogram.

Aiming at the problem of inaccurate calculation of the final idler speed caused by the accumulation of errors when calculating the idler speed step by step, it is proposed to use a 3D convolution network to detect the idler speed end-to-end, and then judge according to the difference between the idler speed and the belt speed The status of the idler can realize the abnormal detection of the idler. In the experiment, a 3D convolutional network was used to extract the spatio-temporal features of the rollers in the video, and the speed of the rollers to be measured was classified according to the extracted feature results. Create a roller image data set for deep learning, and use a variety of data set expansion methods to expand the size of the original data set to 10 times the original size, which improves the detection accuracy of the final model. Finally, five evaluation indicators were used to evaluate the 3D idler speed detection model, and the model evaluation results met the experimental expectations.

Video-based idler abnormality detection realizes automatic detection of idler abnormalities, reduces the workload of inspection workers, and finds abnormal idlers in time. Compared with the detection method that requires a large number of sensors and the abnormal detection method of rollers based on sound signal characteristics, the video-based detection method reduces the difficulty of sensor installation and maintenance, improves the detection accuracy of jammed rollers, and determines the location of abnormal rollers. Reduce the difficulty and workload of secondary positioning of abnormal rollers during maintenance.

[1] 王腾飞. 矿用带式输送机智能化控制系统设计与应用[J]. 煤矿机械, 2020, 41(12):183-186.

[2] 邱明权. 矿用带式输送机托辊健康监测方法研究[D]. 中国矿业大学, 2018.

[3] 倪凡凡. 远程带式输送机托辊故障检测方法研究[D]. 宁夏大学, 2018.

[4] 张文强. 圆管带式输送机传动结构与输送带接触力学特性分析[D]. 湖南科技大学, 2017.

[5] 王亭亭. 矿用带式输送机安全监测系统研究[D]. 中国矿业大学, 2016.

[6] 曹文宇. 矿用带式输送机安全监测系统改造研究[J]. 机械管理开发, 2018, 33(05):69-71.

[7] 乔崇全. 顺槽带式输送机张紧装置的研究[D]. 中国矿业大学, 2014.

[8] 吕鹏飞, 何敏, 陈晓晶, 鲍永涛. 智慧矿山发展与展望[J].工矿自动化, 2018, 44(09):84-88.

[9] 李枫. 计算机与信息技术在智慧矿山中的应用——评《智慧矿山技术》[J]. 矿业研究与开发, 2020, 40(05):164.

[10] 王国法, 王虹, 任怀伟, 赵国瑞, 庞义辉, 杜毅博, 张金虎, 侯刚. 智慧煤矿2025情景目标和发展路径[J]. 煤炭学报, 2018, 43(02):295-305.

[11] W. Pratt Rogers,M. Mustafa Kahraman, Frank A. Drews,Kody Powell,Joel M. Haight,Yaxue Wang,Kritika Baxla,Mohit Sobalkar. Automation in the Mining Industry: Review of Technology, Systems, Human Factors, and Political Risk[J]. Mining, Metallurgy & Exploration, 2019, 36(4).

[12] 邓志远. 无人机在高压输电线路巡视中的应用研究[D]. 华北电力大学(北京), 2015.

[13] 高旭东. 无人机线路巡检安全距离的分析与测量[D]. 太原理工大学, 2019.

[14] 李学民. 矿用巡检机器人关键技术分析[J]. 煤矿机械, 2018, 39(05):71-73.

[15] 张树生, 马静雅, 陆文涛, 裴文良. 矿用带式输送机巡检机器人控制系统设计与实现[J]. 煤矿机械, 2015, 36(07):28-30.

[16] 马静雅,岑强,李军伟.矿用带式输送机巡检机器人系统设计[J].煤炭技术,2016,35(01):249-251.

[17] 王彦, 黄晓东, 黄朝晖. 煤矿带式输送机智能巡检系统的设计与应用[J]. 煤矿机电, 2014(02):39-41.

[18] 曹贯强. 带式输送机托辊故障检测方法[J].工矿自动化, 2020, 46(06):81-86.

[19] 孙维, 刁冬梅.基于φ—OTDR技术的带式输送机托辊故障检测[J]. 工矿自动化,2016,42(08):9-12.

[20] Zhenya Wang, Ligang Yao, Yongwu Cai. Rolling bearing fault diagnosis using generalized refined composite multiscale sample entropy and optimized support vector machine[J]. Measurement, 2020, 156.

[21] L. Mubaraali,N. Kuppuswamy,R. Muthukumar. Intelligent fault diagnosis in microprocessor systems for vibration analysis in roller bearings in whirlpool turbine generators real time processor applications[J]. Microprocessors and Microsystems,2020,76.

[22] Kobayashi Yusuke, Song Liuyang, Tomita Masaru, Chen Peng. Automatic Fault Detection and Isolation Method for Roller Bearing Using Hybrid-GA and Sequential Fuzzy Inference.[J]. Sensors (Basel, Switzerland), 2019, 19(16).

[23] Aleksandra Grzesiek, Radosław Zimroz, Pawel Śliwiński, Norbert Gomolla, Agnieszka Wyłomańska. Long term belt conveyor gearbox temperature data analysis – Statistical tests for anomaly detection[J]. Measurement, 2020, 165.

[24] Jarosław Szrek, Jacek Wodecki, Ryszard Błażej, Radoslaw Zimroz. An Inspection Robot for Belt Conveyor Maintenance in Underground Mine—Infrared Thermography for Overheated Idlers Detection[J]. Applied Sciences, 2020, 10(14).

[25] 谢苗, 朱振, 卢进南. 基于红外图像处理技术的托辊卡阻检测方法[J]. 机械设计与究, 2020, 36(05):152-157.

[26] Meghdad Khazaee,Ahmad Banakar, Barat Ghobadian, Mostafa Mirsalim, Saeid Minaei, Mohamad Jafari, Peyman Sharghi. Fault detection of engine timing belt based on vibration signals using data-mining techniques and a novel data fusion procedure[J]. Structural Health Monitoring, 2016, 15(5).

[27] 韩涛, 胡英贝, 张蕾, 张文涛, 徐振宇. 信息融合技术在托辊轴承故障诊断中的应用[J]. 轴承, 2012(06):57-59.

[28] 张杰, 李浙昆, 翟守忠. 带式输送机组远程集中监控系统的研究与开发[J]. 矿业研究与开发, 2013, 33(06):100-102+117.

[29] 鲍志鹏,沈希忠,韩志威.基于EMD的齿轮箱故障诊断的研究[J].煤矿机械,2015,36(02):282-284.

[30] Zhou Hongyan. Study on Detection and Fault Diagnosis System of Transmission of Coal Cutter Based on Improved BP Neural Network[J]. Chemical Engineering Transactions (CET Journal), 2018, 71.

[31] Bharath Subramani, Magudeeswaran Veluchamy. Quadrant dynamic clipped histogram equalization with gamma correction for color image enhancement. 2020, 45(4):644-655.

[32] 韩少刚. 基于多直方图均衡的图像增强算法研究[D]. 安庆师范大学,2020.

[33] Dhiren R. Patel,Harshit Thakker,M B Kiran,Vinay Vakharia. Surface Roughness Prediction of Machined Components Using Gray Level Co-occurrence Matrix and Bagging Tree[J]. FME TRANSACTIONS, 2020, 48(2).

[34] Sharmila Biswas, Sandeep Singh Solanki. Singer Identification using Autocorrelation Method[J]. International Journal of Recent Technology and Engineering (IJRTE), 2020, 9(4).

[35] 刘洋, 刘晓波, 梁珊. 基于傅里叶分解方法的航空发动机转子故障诊断[J]. 中国机械工程, 2019, 30(18):2156-2163.

[36] 马祥, 杜忠华, 蔡雨, 王鹏飞, 卿志勇. 融合梯度信息的改进中值滤波算法研究[J]. 传感器与微系统, 2021, 40(03):48-51.

[37] 李肖肖, 聂仁灿, 周冬明, 谢汝生. 图像增强的拉普拉斯多尺度医学图像融合算法[J]. 云南大学学报(自然科学版), 2019, 41(05):908-917.

[38] 陈顺, 孟青青, 李登峰. 结合图像增强和改进Canny算子的遥感图像边缘检测[J]. 河南大学学报(自然科学版), 2020, 50(05):623-630.

[39] M Ravi Kumar, J Lakshmi Prasanna, Varanasi Sri Harsha Vardhan, Morla Bhaskar Vystanava Swamulu, Gade Kalyan Kumar, Sreevardhan Cheerla, Chella Santhosh. Image Enhancement for Ultrasound Images Using Sobel Edge Detection[J]. Journal of Critical Reviews, 2020, 7(13).

[40] 罗萍, 胡光宝, 吕霞付, 康健. 基于改进LOG算子的图像增强算法[J]. 微电子学与计算机, 2019, 36(09):26-29

[41] 张会珍, 刘云麟, 任伟建, 刘欣瑜. 人体行为识别特征提取方法综述[J].吉林大学学报(信息科学版), 2020, 38(03): 360-370.

[42] 祁大健, 杜慧敏, 张霞, 常立博. 基于上下文特征融合的行为识别算法[J]. 计算机工程与应用, 2020, 56(02):171-175.

[43] Simonyan K, Zisserman A. Two-stream convolutional networks for action recognition in videos. In: Proceedings of Advances in Neural Information Processing Systems (NeuPS2014), 2014. 568–576.

[44] Varol G, Laptev I, Schmid C. Long-term temporal convolutions for action recognition. IEEE Trans Pattern Anal Mach Intell, 2018, 40: 1510–1517.

[45] Tran D , Bourdev L , Fergus R , Torresani L, Paluri M. Learning Spatiotemporal Features with 3D Convolutional Networks [J]. 2014.

[46] 徐访, 黄俊, 陈权. 基于3D卷积神经网络的动态手势识别方法[J/OL]. 计算机工程:1-9[2021-02-16].

[47] 何景琳, 梁正友, 孙宇, 刘德志.结合C3D与光流法的微表情自动识别[J]. 计算机系统应用, 2021, 30(01):221-227.

[48] 刘宇琦. 基于深度学习的托辊异常检测方法研究[D].西安科技大学,2020.