Abstract: With the development of computer and sensor technology, a large number of process data during manufacturing is collected. Extracting the knowledge and information from the data is an important method to improve the product quality. The clustering method with process data is used to acquire the production status, thus for process diagnosis or enhancing the focal points of the quality inspect. Spectral clustering is one of the advanced methods, where the Euclidean distance is used as the common similarity measure, which can only extract the features of the spherical or hyperspherical distribution data but can not express the complex distribution data. In this paper geodesic distance is introduced to spectral clustering, which is used to do the production state clustering. The benchmark data, Tennessee Eastman process data are used for model validation. As a result the proposed method reduces the parameter sensitivity and has better performance on clustering to acquire the production state, compared with the Euclidean distance.

Key words: geodesic distance, spectral clustering, production state analysis, clustering analysis

摘要： 随着计算机和传感技术的发展，大量生产过程数据被记录。提取数据中的知识信息是提高产品质量的重要手段，通过聚类分析可以了解生产状态，进行生产故障诊断或有针对性的质量检测，谱聚类是较为先进的聚类方法，而传统的谱聚类中使用欧式距离作为相似性的度量，但欧式距离只能反映数据空间分布为球形或超球形的结构特性，难以刻画复杂数据分布特性，将测地距离引入谱聚类中，并应用于生产过程状态的聚类分析中，分别利用标准数据、TE生产过程数据对方法的有效性进行验证，结果表明测地距离谱聚类方法可以降低参数的敏感性，且具有更优的聚类结果，可以更加有效了解生产过程状态。

关键词: 测地距离, 谱聚类, 生产状态分析, 聚类分析