钢丝绳芯带式输送机是目前大多数煤矿的主运输设备，在现代化煤矿生产中发挥着极其重要的作用。钢丝绳芯输送带使用过程中因发生接头位移、断绳、断丝和疲劳等缺陷导致输送带断裂事故频发，造成了人员伤亡和重大经济损失。为了克服目前输送带缺陷人工目测方法的缺点和不足，论文通过研发煤矿钢丝绳芯输送带弱磁在线检测研究平台，运用理论与实验相结合的方法对缺陷磁场空间分布规律和弱磁检测信号智能识别理论和算法进行了深入研究。 针对目前煤矿生产中钢丝绳芯输送安全在线检测这一亟待解决的技术难题，借助于电磁无损检测与评价技术、计算机技术、信号处理技术、变频控制技术和机电一体化技术，成功地开发了输送带缺陷弱磁在线智能识别系统，实现了煤矿钢丝绳芯输送带缺陷弱磁检测的自动化，提高了输送带缺陷检测的可靠性和效率，对确保输送带安全可靠运行具有重要意义。 研究了钢丝绳芯输送带断绳和接头位移的磁场空间分布问题，运用磁偶极子模型建立了断绳和接头的磁场空间分布数学模型。通过数学模型构建了接头位移和断绳缺陷样本，分析了接头磁场沿输送带长度方向(x方向)和垂直输送带方向(y方向)磁场空间分布随接头位移变化规律。研究结果表明：y方向磁场空间分布的波峰与波谷间距可以准确反映接头位移的变化，为接头位移定量评价提供了依据。 研究了在煤矿强噪声背景下的钢丝绳芯输送带缺陷弱磁检测信号降噪问题，提出了一种改进阈值小波的变步长LMS自适应滤波算法。该算法不仅融合了小波变换和自适应滤波的优点，而且通过对小波阈值处理函数和LMS自适应滤波步长的改进获得了比小波和自适应滤波更好的降噪性能。通过对多种降噪算法的比较分析表明：该算法对输送带缺陷信号中的非平稳噪声的降噪具有良好效果，有效地提高了信噪比。 研究了输送带多传感器信息融合特征提取和特征约简问题，提出了多传感器时域特征和小波包能量特征加权融合的特征提取算法和基于属性数据标准差的改进邻域粗糙集特征约简算法。首先，根据弱磁传感器检测的缺陷信号提取时域特征和小波包能量特征；其次，对提取的缺陷特征进行加权融合；最后，对缺陷特征进行约简。实验结果表明：该算法可以有效地提取本质特征向量。 研究了钢丝绳芯输送带多类缺陷信号分类问题，提出了一种基于改进遗传算法优化的模糊二叉树支持向量机多类分类算法。实验结果表明：该算法可以较快地获得支持向量机的一组最优参数 和 ，分类精度高，训练速度快，支持向量个数少，并且对于含有噪声数据的分类比常用支持向量机多分类算法具有更高的分类精度和更少的支持向量。 研究了缺陷的定位和定量分析问题，提出了一种基于特征间距的接头编号方法，对接头位移进行定位，运用多个煤矿现场检测接头数据验证了该方法的有效性。对于接头位移的定量分析，运用多个垂直传感器的波峰与波谷之间的间距加权平均值作为接头长度，并根据接头长度的变化量对接头位移进行定量，实验发现该方法定量精度较高。提出了基于多传感器融合的分区定位方法，对非接头区缺陷进行宽度方向定位，以相邻4个传感器为一组对输送带宽度进行分区，然后在分区内根据相邻水平传感器幅值信息进行融合定位，结果表明：该定位方法的定位误差为1/2个区，并且发现在传感器之间互不影响的条件下，对传感器密集布置可以提高缺陷宽度方向定位精度。

Presently, steel cord belt conveyor is main transportation equipment of most coal mines. It plays an extremely important role in the modern production of coal mine. Steel cord conveyor belt may occur to defects such as broken rope, broken wire, fatigue, joint displacement and etc in the process of using. The defects lead to frequent accidents of conveyor belt rupture which causes to casualties and significant economic loss. In order to overcome shortcomings and deficiencies of visual inspection, weak magnetic online testing research platform was developed. Magnetic field spatial distribution laws of defects and weak magnetic testing signal intelligent recognition theories and algorithms were deeply studied by the method of combining theory and experiment. For the safe online testing problem of steel cord conveyor belt in coal mine, intelligent recognition system of steel cord conveyor belt defects was developed by electromagnetic nondestructive testing and evaluation technology, computer technology, signal processing technology, frequency conversion control technology and mechatronics technology. The system realized automation of defects weak magnetic testing and improved reliability and effiency of defects testing. It has great significance for ensuring safe and reliable operation of conveyor belt. The magnetic field spatial distribution problems of broken rope and joint displacement were studied, magnetic field spatial distribution mathematical model of broken rope and joint were founded based on magnetic dipole model. Defects samples of joint displacement and broken rope were achieved by using the mathematical model. The variation laws between joint magnetic field spatial distribution along conveyor belt length direction(x direction) and vertical conveyor belt direction(y direction)and joint displacement were analyzed. The research results showed that spacing of peaks and troughs for y direction magnetic field spatial distribution can accurately reflect the changes of joint displacement. It provides the basis for quantitative evaluation of joint displacement. The problem of weak magnetic signal filtering was studied in the strong noise background of coal mine, a variable stepsize LMS adaptive filtering algorithm based on improved threshold wavelet transform is presented. The method not only combines advantages of wavelet transform and adaptive filtering, but also it has better filtering performance than wavelet transform and adaptive filtering by improved wavelet threshold processing function and LMS adaptive filtering step-size. By a variety of filtering algorithm studied, it showed that the new algorithm had good filtering effect to non-stationary noise of defects signal and effectively improved the signal to noise ratio. The problems of feature extraction and reduction for defects signal with multi-sensor information fusion were studied, a weighted fusion feature extraction algorithm of multi-sensor time-domain feature and energy feature of wavelet packet and a modified neighborhood rough set feature reduction algorithm based on attribute data standard deviation were presented. First, the time-domain feature and energy feature of wavelet packet for defects signal were extracted; Secondly, defects features were fused with weight. Finally, defects feature was reduced. The experimental results showed that the new algorithm can effectively extract nature feature vectors of defects signal. The multi-class classification problem of defects signal was studied. A multi-class classification algorithm based on improved genetic algorithm optimized fuzzy binary tree support vector machines was presented. The experimental results showed that the algorithm can quickly obtain optimal parameters and of support vector machine. It has the advantages of high classification accuracy, high training speed and less support vectors, as well as better classification accuracy and fewer support vectors than usual support vector machines multi-classification algorithm for the data classification with noise. The problems of defects location and quantitative analysis were studied. A joint number location method based on characteristic spacing was presented. The validity of the method was verified by large number of coal mine site joint testing data. For the joint displacement quantitative analysis, weighted average spacing between peaks and troughs of vertical sensors was used as joint length and variation of joint length was equaled to joint displacement. The experimental results showed that the algorithm had good repeatability and high quantitative accuracy. For location analysis of non-joint area defects, regional location method based on multi-sensor fusion was put forward. First, the conveyor belt width was separated by adjacent four sensors.Then,fusion location was executed by amplitude of adjacent horizontal sensors in one area. The results showed location error of the method for 1/2 area. It is founded that intensive layout of sensors can improve location accuracy of defects width direction under the condition of independently for each other between sensors.