【作者】 夏战国；

【导师】 夏士雄；

【作者基本信息】 中国矿业大学 ， 计算机应用技术， 2013， 博士

【摘要】 随着科学技术的进步和工业需求的发展，提升机一方面不断向大型、复杂、高速、高效的方向发展，另一方面却又面临更加苛刻的工作和运行环境。提升机轴承作为矿井提升设备的关键装备，是整个提升系统的重要组成部分和核心关键部件，同时也是最易出现故障的薄弱部件，它的运行状态好坏直接影响到整个系统的运行。一旦提升机轴承部件发生故障，就可能破坏整台设备甚至影响整个生产过程，造成巨大经济损失，还可能导致灾难性的人员伤亡和形成严重的社会影响。因此，如何有效评测提升机轴承的运行性能，研究制定设备维修计划，借助智能信息处理的手段实行设备智能维护，确保设备安全可靠运行，避免或减少各种恶性事故突发，对煤矿企业安全生产、减灾救灾和煤炭工业的健康持续发展来说是当前迫切需要解决的问题。本论文的主要研究工作包括：（1）针对设备性能退化数据中的高维特性会导致后续评测算法效率低下的问题，提出了一个基于谱回归（SR）的轴承性能特征提取方法。该方法首先通过标记和非标记样本构造仿射近邻图，以揭示给定数据的内在结构信息，随后根据所学得的响应值，再用普通的回归方法学习嵌入函数，最后采用最小二乘法获得最佳映射方向，避免了计算特征空间稠密矩阵的问题。通过多组公测轴承实验数据和提升机轴承实验数据证实了SR方法比PCA、FA以及其他常用的流形学习技术更有效。（2）针对有监督的机器学习方法在训练评估模型时需要大量正确标记的训练样本以及训练样本类不均衡的问题，提出了一个基于半监督高斯过程的分类算法。该算法首先根据类不均衡数据的特性进行数据预处理，然后利用少量的标记数据进行高斯过程分类训练，选取预测概率置信度最高的未标记数据，向该未标记数据注入合理的类标记信息，并且自动地将新标记过的数据样本加入到原有的训练集中，用扩充后的训练集再次进行高斯过程分类，最后采用自训练迭代执行，直至构造出性能最优的半监督高斯过程分类器。通过仿真数据和公测轴承数据进行了验证，并成功用于提升机轴承性能退化评估实验中。（3）针对设备性能退化数据中自然存在的时序性难以用于标准的高斯过程回归的问题，提出了一个基于隐马尔可夫高斯过程的预测算法。该算法很好地结合了HMM的时序预测特性和高斯过程回归的核方法预测优势，建立了一个HMMGP模型，将每一个设备性能退化状态对应成一个高斯过程，利用EM算法学习模型参数，并使用共轭梯度法求得似然函数极大时的参数，为了避免求逆矩阵运算，采用Cholesky分解法降低了算法复杂度。论文用多组轴承全寿命实验数据进行了验证，并成功用于提升机轴承性能退化预测实验中，与其他预测方法进行了对比，表明HMMGP预测模型具有更高的预测精度。本文研究的内容涉及到了提升机轴承性能特征提取、轴承性能退化评估和性能退化预测三方面的问题，主要采用了高斯过程机器学习方法及相关改进算法。在研究了相关领域前期工作的基础上对已有的算法进行了分析和改进，使用了轴承故障诊断领域中常用的公测数据集进行了实验，并与相关算法进行了性能对比，验证了本文提出的算法的有效性，最后在提升机轴承性能退化评测中对文中所提出的算法进行了应用。应用结果表明，本文的研究成果在提升机轴承性能评测中取得了不错效果，对提高煤矿生产及设备可靠性具有一定的指导意义。更多还原

【Abstract】 The key equipment of mine hoisting is the hoist bearing, it is also the importantpart of the lifting system and the core components; meanwhile, it is the weak part ofthe system and easy to cause fault; thus its operating status directly affects theoperation of the whole system. Therefore, how to effectively assess the operationalstatus of hoist bearing, how to study and formulate equipment maintenance program,how to use the means of intelligent information processing to implement intelligentmaintenance of the equipments, which is an urgent problem to be solved to make surethe operation of the equipments is safe and reliable, so that a variety of unexpectedfatal accidents can be avoided or reduce. In the field of coal production, it is also animportant topic to make the process of production safely and the coal industryhealthily.The main research work is elaborated as following:(1) In order to improve the efficiency of assessment and prediction algorithmwhen dealing with the data degradation in the high-dimensional feature space, amethod, which is based on the spectral regression, is proposed to capture the featuresof the hoist bearing. Through multiple sets of experimental data of bearing, it isconfirmed that the method of spectral regression is more effective than PCA, FA orthe other common manifold learning techniques.(2) As the methods of supervised learning normally need a larger number ofcorrectly per-labeled training examples and the distribution of training examples isimbalance, a method of Semi-supervised Gaussian process is proposed in the field ofassessing the degradation of the hoist bearing. The effectiveness of such method isproven through several experiments in practice.(3) To improve the efficient of Gaussian process when dealing with equipmentperformance degradation data, a method of Hidden Markov Gaussian process isproposed in the field of assessing the degradation of the hoist bearing. The method iswell combined the kernel-based Gaussian process regression with the characteristicsof HMM in predicting the performance degradation data and can be used to build up aHMMGP model. The comparing tests are carried out by using the some sets ofexperimental data of full life of the bearings. HMMGP prediction model has higherprediction accuracy.The contents in this thesis include capturing the features of the hoist bearing, assessing and predicting the degradation of the bearing by using the Gaussianprocesses learning and the related improved methods. The proposed algorithm is usedin a real case to assess the degradation of the hoist bearing, the result of theapplication shows that the effectiveness of the new algorithm when it is used to assessthe degradation of the hoist bearing, and it has great influence on improving thereliability of mining equipment and process of the mining production.更多还原