基于自适应Mean Shift的结构健康状态监测技术研究

发布时间：2018-04-13 22:38

本文选题：自适应Mean + Shift　；参考：《长安大学》2013年硕士论文

【摘要】：随着科学技术的发展，机械设备越来越复杂，自动化水平越来越高，机械设备在现代工业生产中的作用和影响越来越大，，与其有关的费用越来越高，机器运行中发生的任何故障或失效不仅会造成重大的经济损失，甚至还可能导致人员伤亡。因此，应该及时地对设备故障状态进行监测，使之安全经济地运转。本文以滚动轴承为研究对象，研究了基于自适应Mean Shift聚类算法的机械结构健康状态监测技术。研究了信号的预处理方法：时域指标、频域指标和小波包变换。实验表明：当滚动轴承出现故障时，时域、频域指标都会发生变化，且不同类型损伤和损伤程度不同时，时域和频率指标有明显差别；另外，不同类型损伤的振动信号经小波包变换分解后，其能量分布也表现出不同的特征；因此，提取振动信号的时域、频域和小波包指标可以降低振动信号的维数，有效地描述不同类型的故障状态。研究了基于能量熵的健康状态监测方法，实验表明：小波包能量熵可以有效地鉴别故障状态和损伤程度，可以用来监测滚动轴承健康状态的变化历程。论述了Mean Shift算法的原理，通过实验研究了核函数、核半径以及阈值对聚类算法的影响。核函数影响聚类分析的准确率及算法的迭代次数，对于每一个核函数，聚类算法都存在一个合理的核半径区间，当核半径超出该范围时，聚类的准确率会降低；另外，阈值越小，算法的聚类效果越好。论述了自适应Mean Shift算法（Adaptive Mean Shift,AMS）的原理，通过实验证明了核函数、初始核半径以及迭代次数对聚类算法的影响。使用高斯核函数时，核半径初值的选择对聚类影响较大，使用Epanechnikov核函数时，核半径初值对聚类影响较小，与前者相比，它的分类准确率较低；另外，增大迭代次数，可以改善聚类效果。与MeanShift算法相比，AMS算法具有较好的聚类效果。提出了基于自适应Mean Shift质心偏移的结构健康状态监测方法，该方法将无损伤状态的质心作为基准，用某一状态的质心与基准质心的偏移量判断结构是否发生损伤以及损伤程度。实验表明：与基准质心距离越远，结构的损伤程度越严重。因此用质心偏移量可以有效地评估结构的健康状态。
[Abstract]:With the development of science and technology, machinery and equipment are becoming more and more complex, the level of automation is becoming higher and higher, and the role and influence of machinery and equipment in modern industrial production are becoming greater and greater, and the costs associated with them are becoming higher and higher.Any failure or failure in the operation of the machine will not only cause great economic losses, but also may lead to casualties.Therefore, the equipment failure condition should be monitored in time to make it run safely and economically.In this paper, based on the adaptive Mean Shift clustering algorithm, the health state monitoring technology of mechanical structure is studied.The signal preprocessing methods: time domain index, frequency domain index and wavelet packet transform are studied.The experimental results show that when the rolling bearing failure occurs, the time domain and frequency domain indexes will change, and different types of damage and damage degree will have obvious difference between time domain and frequency index.The energy distribution of the vibration signals with different types of damage is decomposed by wavelet packet transform, so the dimension of vibration signals can be reduced by extracting the time-domain, frequency-domain and wavelet packet indexes of the vibration signals.Effectively describes different types of fault states.The method of monitoring health state based on energy entropy is studied. The experimental results show that the wavelet packet energy entropy can effectively identify the fault state and damage degree, and can be used to monitor the health state of rolling bearing.The principle of Mean Shift algorithm is discussed, and the effects of kernel function, kernel radius and threshold on clustering algorithm are studied experimentally.Kernel function affects the accuracy of clustering analysis and the number of iterations of the algorithm. For each kernel function, the clustering algorithm has a reasonable kernel radius interval. When the kernel radius exceeds this range, the clustering accuracy will be reduced.The smaller the threshold, the better the clustering effect of the algorithm.The principle of adaptive Mean Shift algorithm is discussed. The effects of kernel function, initial kernel radius and number of iterations on the clustering algorithm are proved by experiments.When Gao Si kernel function is used, the choice of initial value of kernel radius has a great influence on clustering. When Epanechnikov kernel function is used, the initial value of kernel radius has less effect on clustering, compared with the former, its classification accuracy is lower; in addition, the number of iterations is increased.The clustering effect can be improved.Compared with MeanShift algorithm, it has better clustering effect.A method of structural health state monitoring based on adaptive Mean Shift centroid migration is proposed. The centroid of non-damaged state is used as the reference, and the deviation of centroid and reference centroid of a certain state is used to judge whether the structure is damaged or not and the degree of damage.The experimental results show that the longer the distance from the reference centroid, the more serious the damage degree of the structure is.Therefore, the centroid deviation can be used to evaluate the health status of the structure effectively.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TH165.3;TH133.3

【参考文献】