【作者】 李万召；

【导师】 闫纪红；

【作者基本信息】 哈尔滨工业大学 ， 机械电子工程， 2008， 硕士

【摘要】 随着时代发展,工业生产正逐渐向系统化、协同化方向发展,生产设备之间的联系日趋紧密,生产系统日趋复杂。其中,有些设备在整个系统中作用重大,有些设备则十分昂贵。如果这些关键设备的性能发生劣化,可能会导致系统瘫痪或者严重的经济损失。鉴于此,及时准确的性能评价是非常必要的。本文主要研究了设备非平稳信号特征提取和特征选择,及基于支持向量机的设备性能评价方法。在特征提取方面,本文分别采用了连续小波变换、离散小波重构、小波包分解求频带能量等多种方法,经过对结果的深入分析和比对,最终选定了小波包分解求频带能量为最优方法。值得一提的是,本文在研究中发现了信号不同频段变化趋势的相似性,将成组技术思想引入信号分析领域,并对成组理论进行了深入分析,提出了成组技术以广义夹角余弦为度量的非线性矢量空间理论框架,并设计和实现了一种新的成组算法,收到了良好的应用效果。在特征选择方面,本文提出了未知信号的两指标联合选择方法,对模型的预期性质进行了深入探讨,并最终构建了一个满足该性质的数学模型。实际验证结果表明,本模型的选择能力完全达到了预期目标。除此之外,该模型对于类似的选择模式可推广到多指标选择领域,具有一定的普适性。为了特征融合算法的研究更加直观简便,本文还采用了主成份分析对信息做了进一步的压缩,将特征向量降维到可视空间内。在特征融合方面,本文对支持向量机的基本原理做了精炼系统的阐述,并对各种类型支持向量机算法做了大量的仿真实验,积累了丰富的实际经验。经过分析比对,最终选定两种特征融合方法:“一对多”分类器算法和支持向量回归机算法。其中,“一对多”分类器算法用于设备的阶段性能评价,支持向量回归机用于设备的连续性能评价。实验最终结果显示,该两种融合结果在泛化能力和分析精度上都非常优秀。最后,本文采用LabVIEW软件建立了一套完整的设备性能评价实验系统。其中,实验对象是一台搬运机械手,经过可行性分析及对实际硬软件条件考虑,确定监测气路密封性和传动系统灵活性两个方面的性能。经过对历史数据的分析,确定了相关的特征提取方法和特征融合方法,并获得了特征融合的数学模型。最终,数据采集、特征提取、特征融合、输出与存储等全部集成在以MATLAB计算引擎为后台的LabVIEW在线性能评价系统中,实际效果良好。更多还原

【Abstract】 With the development of the times, industrial production is gradually tending towards to be systematic and coordinated; relationships between equipments are increasingly more close and complex in the production system. Therefore, some equipments play key roles in the whole system, some of them are very expensive. The system may result in paralysis or serious economic losses if the performance of these key equipments degrades. In view of this, timely and accurate performance evaluation is essential.This thsis studied the feature extraction and selection methods for non-stationary signals and the evaluation methods of equipment performance based on support vector machines (SVM).In the field of feature extraction, adopted a number of methods, including Continuous Wavelet Transform, Discrete Wavelet Transform, Wavelet Reconstruction, Wavelet Packet Transform, etc were applied in this thesis. The Wavelet Packet Transform for the band-energies is selected as the optimal method after a series of parametric studies.Particularly, the similarity of signal changing trends in different bands is found. The Group Technology is introduced into signal analysis field in this thesis as well as. A group technology theoretical framework based on nonlinear vector space with the measurement of generalized angle cosine is constructed. A new algorithm for grouping is designed and applied. The simulation results are convictive.In the field of feature selection, the two indexes option-method for unknown signals was applied in this thesis. The expected characters of the model were studied. A mathematical model meeting the requirements was eventually constructed. The simulation results show that the selective ability of the model fully agreed the expected goals. In addition, the model can be extended to the field of multi-indexes selection with similar model. In other words, it has a universal application. In order to study the back-end of algorightm more intuitively and simply, PCA was applied to compress information further. The dimensions of feature vectors were reduced into the visual space. In the field of feature fusion, the basic principles of Support Vector Machine were described refinedly and systemically in this thsis. Simulation experiments for various types of SVM algorithms were performed, practical experiences were accumulated as well as. After being analyzed and contrasted, two feature fusion methods were selected: "one-to-others" classification algorithm and SVM regression algorithm. In which "one-to-others" classification algorithm equips is for the stage performance evaluation and SVM regression algorithm is for continual performance evaluation. Experimental results show that, the two feature fusion methods are both perfect at generalization ability and analysis accuracy.Finally, a complete set of equipment system for performance evaluation was created by LabVIEW software. The object studied is a conveyer mechanical hand. Through feasibility analysis and deep advisement of actual hardware and software, the hermetical performance of pneumatic system and agility of transmission system were determined. By analyzing the history data, the relevant feature extraction and fusion methods were determined. The mathematical models were abtained. All modules (such as data acquisition, feature extraction, feature fusion, output and storage, etc) are integrated into the LabVIEW online performance evaluation system with background of MATLAB computing engine. The performance of the system is promising.更多还原