基于混合杂草算法—神经网络的转子故障数据分类方法研究

发布时间：2018-02-22 02:02

本文关键词： 转子系统信息熵混合杂草优化算法神经网络　出处：《兰州理工大学》2012年硕士论文　论文类型：学位论文

【摘要】：近年来，随着现代机械设备的大型化、复杂化、自动化和连续化，开展机械设备的故障诊断技术的研究具有重要的现实意义。目前，国内外学者在此方面做了大量的工作，使得相关的理论与应用取得的迅猛的发展。机械故障诊断是通过研究故障与征兆之间的关系来判断设备故障的，而故障与征兆之间表现出的非常复杂的非线性关系，很难用数学模型加以精确的描述，给机械的故障诊断带来很大的不便。人工神经网络是一种重要的人工智能行为，是一个非线性计算系统，可以实现故障与征兆之间复杂的非线性映射关系，因此在机械故障诊断领域得到了极大的应用潜力。本文构建的混合杂草优化算法优化BP神经网络预测模型，以转子试验台模拟的大量的故障数据为支持，采用信息熵方法来定量的对故障数据进行特征提取，混合的杂草优化算法优化神经网络结构。主要工作内容和研究成果如下：（1）在转子实验台上模拟了四种典型故障，对信息熵的性质和时域的奇异谱熵、频域的功率谱熵、时频域的小波能谱熵和小波空间谱熵进行了较为系统的研究和探讨。（2）以四类谱熵为原始数据，对数据进行归一化处理，并建立训练样本库和测试样本库。（3）在分析遗传算法、粒子群算法优点的基础上，将遗传算法中的交叉算子、粒子群算法的矢量操作引入IWO，提出了HIWO。（4）建立了HIWO优化BP神经网络模型，由HIWO算法训练BP网络训练的初始最优权值和阈值，然后在训练好的BP神经网络中对测试样本进行预测，并且与遗传算法、粒子群算法及IWO优化的神经网络进行了对比分析。（5）基于HIWO算法流程开发了一套MATLAB GUI的转子故障诊断系统，，子系统一实现对振动信号的消噪分析，频谱分析，轴心轨迹分析等；子系统二实现熵值数据的归一化；子系统三实现四种算法优化神经网络的初始权值和阈值；子系统四根据样本特点对分类器进行参数寻优，实现对未知故障的判别，实验证明了该系统的有效性。
[Abstract]:In recent years, with the large-scale, complex, automation and continuity of modern mechanical equipment, it is of great practical significance to carry out the research on fault diagnosis technology of mechanical equipment. At present, many scholars at home and abroad have done a lot of work in this field. Mechanical fault diagnosis is based on the study of the relationship between the fault and the symptoms to judge the fault of the equipment, and the relationship between the fault and the symptoms shows a very complex nonlinear relationship. It is difficult to describe accurately by mathematical model, which brings great inconvenience to the fault diagnosis of machinery. Artificial neural network is an important artificial intelligence behavior and a nonlinear computing system. The complex nonlinear mapping relationship between fault and symptom can be realized, so it has great application potential in the field of mechanical fault diagnosis. In this paper, a hybrid weed optimization algorithm is constructed to optimize BP neural network prediction model. Based on a large number of fault data simulated by the rotor test-bed, the information entropy method is used to quantitatively extract the fault data. The hybrid weed optimization algorithm is used to optimize the neural network structure. The main work and research results are as follows:. In this paper, four typical faults are simulated on the rotor test bench. The properties of information entropy and singular spectral entropy in time domain, power spectrum entropy in frequency domain, wavelet spectrum entropy in time-frequency domain and wavelet space spectral entropy in time-frequency domain are systematically studied and discussed. (2) taking four kinds of spectral entropy as raw data, the data are normalized, and the training sample database and test sample database are established. On the basis of analyzing the advantages of genetic algorithm and particle swarm optimization algorithm, the crossover operator and the vector operation of particle swarm optimization algorithm in genetic algorithm are introduced into IWO.The HIWO is proposed. (4) the HIWO optimized BP neural network model is established. The initial optimal weights and thresholds of BP network training are trained by HIWO algorithm, and then the test samples are predicted in the trained BP neural network and compared with genetic algorithm. Particle swarm optimization (PSO) and IWO neural network are compared and analyzed. 5) based on HIWO algorithm flow, a rotor fault diagnosis system based on MATLAB GUI is developed. Subsystem 1 realizes noise reduction analysis, spectrum analysis, axis locus analysis, etc. Subsystem 2 realizes normalization of entropy value data. Subsystem 3 implements four algorithms to optimize the initial weights and thresholds of neural networks, subsystem 4 optimizes the classifier parameters according to the characteristics of samples, and realizes the identification of unknown faults. The experiment proves the effectiveness of the system.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3;TP183

【参考文献】