|
| 基于小波包和聚类算法的滚动轴承故障检测研究 |
| Research on Fault Detection of Rolling Bearing Based on Wavelet Packet and Clustering Algorithm |
| |
| DOI：10.16018/j.cnki.cn32-1650/n.202301012 |
| 中文关键词: 故障检测 轴承 小波包 聚类算法 |
| 英文关键词: fault detection bearing wavelet packets clustering algorithms |
| 基金项目: |
|
| 摘要点击次数: 838 |
| 全文下载次数: 871 |
| 中文摘要: |
| 针对Kmeans算法在滚轴故障检测中k值需要人工设定以及初始聚类中心的随机选取问题，提出I-Canopy-Kmeans算法的故障检测方法对其进行优化。该算法在初始聚类中心的随机选取方面，使用“最远最近”的原则，即在获取n个Canopy时，任意两个Canopy中心点之间的距离应该尽可能远，且第n个Canopy中心点应该是其他数据点与前面n-1个中心点最远距离中最小的一个；在阈值选取方面，使用欧氏距离求出所有数据点的均值点，再计算均值点到所有数据点的距离，并用L1和L2分别表示最远距离和最近距离，然后将(L1+L2)/2 赋值给阈值T1、(L1+L2)/3赋值给阈值T2。实验结果表明，与传统Kmeans算法相比，I-Canopy-Kmeans算法的各项评价指标均有提高，其中IAR提高最多，达到了40. 01%。 |
| 英文摘要: |
| The Aiming at the problem that k value needs to be set manually and the initial clustering center needs to be randomly selected in the roller fault detection of Kmeans algorithm, a fault detection method of I-Canopy-Kmeans algorithm is proposed to optimize it. This algorithm uses the principle of "farthest closest" in the random selection of initial clustering centers, that is, when obtaining n Canopies, the distance between any two Canopy center points should be as far as possible, and the nth Canopy center point should be the smallest of the farthest distances between other data points and the previous n-1 center points. In terms of threshold selection, Euclidean distance is used to calculate the mean point of all data points, and then the distance from the mean point to all data points is calculated. L1 and L2 are used to represent the farthest and closest distances respectively. Then, (L1+L2)/2 is assigned to threshold T1, and (L1+L2)/3 is assigned to threshold T2. The experimental results show that, compared with the traditional Kmeans algorithm, the I-Canopy-Kmeans algorithm has improved in all evaluation indexes, among which IAR has improved the most, reaching 40. 01%. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|