【作者】 任芳；

【导师】 熊诗波； 杨兆建；

【作者基本信息】 太原理工大学 ， 机械电子工程， 2003， 博士

【摘要】 本文对基于多传感器数据融合技术的煤岩界面识别的理论与方法进行了较为系统的研究。 煤岩界面识别系统能使采煤机具有自动追踪煤岩界面的能力，可靠的识别系统在经济效益和安全作业两方面都具有突出的优点，它能提高煤层的回采率；降低煤中的矸石、灰份和硫的含量；提高采煤作业效率；减轻设备磨损；减少设备维修量和停机时间；降低了空气中的岩尘含量，并可使作业人员远离危险工作面，是实现采煤自动化的关键设备之一。目前各研究方法所使用的都是用单类型传感器进行辨识，由于每个传感器都有其特定的工作精度与适用范围，因此用单传感器拾取信号有一定的局限性，传感器本身的故障与失灵等都会造成误判。因此为了使煤岩界面识别更具有可行性、准确性、可靠性，适应性，能推动煤岩界面识别技术有实质性的突破，为了能给采煤机滚筒自动调高系统产品化提供设计依据，研制开发多传感器辨识系统成为必要。鉴于此提出采用多类型传感器拾取采煤机截割力响应信号并进行多信号特征提取与数据融合的煤岩界面识别方法。方法避免了信号的传输问题，具有可操作性；不受地质条件、采煤工艺的限制，容易推广；采用多信息融合技术，提高了识别正确率，可靠性强。 本文叙述了采煤机的工作原理、分析了采煤机滚筒的受力情况。得出：由于采煤机滚筒在切割煤岩的过程中，随着截割介质(煤或岩)的变化，滚筒的阻力矩、径向作用力以及由截割力激励所产生的结构直线振动响应和滚筒扭转振动响应都要随之发生变化。这些变化蕴涵着截割对象的信息，监测这些参数变化，经过恰当的信号处理与多传感器数据融合(多物理效应的分析与综合)，可实现切割过程煤岩界面识别。 太原理工大学博士学位论文鉴于此通过分析提出，拾取调高油缸压力信号、摇臂振动状态信号、截割电机的电流信号、滚筒轴的扭矩信号、滚筒轴的扭振信号等作为多传感器融合的信息源。 本文建立了采煤机煤岩界面识别物理模拟系统，包括介质模拟和采煤机牵引一切割机构的模拟，研制了试验控制系统，讨论并确定了试验方案。通过试验拾取了大量试验数据，为数据处理准备了信息源。 鉴于小波分析方法的诸多优点，在用经典谱分析的基础上，本文采用小波包技术对信号进行特征提取。通过基于小波包分解的能量分布提取方法能够确定各传感器信号的敏感频段，提取出各特征值。小波包能量法进行特征提取，完成了从模式空间到特征空间的转换，把并不能揭示样本实质的元素过滤掉，使特征的维数大大压缩，获取了最能揭示样本属性的特征量，为数据融合提供了可靠而准确的特征级数据。 数据融合作为一门跨学科的综合信息处理技术，显示了进行煤岩界面状态识别的强大优势。本文分析了利用模糊与神经网络集成技术进行信息融合的优越性并构建了二级模糊神经网络融合系统;针对煤岩界面识别的具体问题，确定了模糊神经网络的具体结构和训练算法;对于网络的输入、输出进行了模糊化处理并对算法进行了修正。结果表明可用试验数据对神经网络、模糊神经网络和二级模糊神经网络进行训练和仿真。基于二级模糊神经网络的数据融合能够进行状态识别并具有较高的识别精度，识别精度大于90%;基于多传感器数据融合的煤岩界面状态识别的方法是可行的。更多还原

【Abstract】 In this thesis, the theory and method of coal-rock interface recognition(CIR) based on multi-sensor data fusion technique are investigated systematically.CIR can make the shearer have the ability to automatically trace the coal-rock interface. It cannot only contribute to mine automation and high efficiency, but also reduce the content of rock and the other mineral that must be removed in the process of coal beneficiation. It is one of the key techniques of mining automation. The present methods collect information by the single sensor only. Because every single sensor has its own special certain precision and application range, collecting information by the single sensor is limited to a certain area and mistakes identification is sometimes unavoidable. Researching and developing an identification system uses multi-sensor is a tendency in order to provide design base for the height adjustment system. Due to the above consideration, the paper has proposed a totally new coal-rock interface recognition method based on multi-sensor data fusion technique with multi-sensor collecting response signal of cutting force. The method avoids the problem of signal transmission, has no limits of zymurgy condition and mining process, can enhance precision and accuracy greatly with multi-signal data fusion technique.The principle of shearer and the force of drum are analyzed. It is concluded: during the period of cutting coal and rock, drum resistance torque, radius acting force and responses of structure straight-line vibrating and drum twist vibrating vary with the variation of shearing media. These variations contain the information of shearing media, monitoring these data changes, with suitable signals process and multi-sensor data fusion(multi physical effects analysis and comprehensive), then coal rock interface recognition canbe realized. Through analyzing, oil pressure signal, arm vibration signal, motor cutting current signal, torsional moment and torsional vibration signals of drum shaft are collected as the information resources of multi-sensor.CIR testing system is established, and the model shearer and model coal wall are designed and manufactured. The control system is manufactured, and test scheme is decided. Through the test, volumes of testing data that are taken as information resource have collected.Feature extraction is carried out with wavelet packet technique because of its advantages. The sensitive frequency bands of signals are judged and their features can be extracted by the energy method. Feature extraction with wavelet packet technique finishes the conversion of pattern space to extraction space, filtrates the factors that cannot reveal sample essentials. It compresses greatly feature dimensionality, acquires the characteristic variables which can reveal sample attributive. It provides correct and reliable feature level data for data fusion.Data fusion, as a cross synthetic information process technique, presents power predominance of carrying out coal-rock interface recognition. The paper probes the advantage of information fusion with the fuzzy neural network(FNN) technique, constructs the two-level FNN. For the CIR problem, the structure and training algorithm of FNN are decided, the input and output of FNN are fuzzed and the algorithm is modified. The results show that the test data can be used to train and simulate with the neural network, FNN, two-level FNN. System adopted FNN technique with the two-level structure can carry out the CIR and the system adopted the modified algorithm has the higher identification ratio which is more than 90%. The method of CIR based on multi-sensor data fusion technique is feasible and more reliable.更多还原