Development and Application of a New Spectral Clustering Algorithm

The changes in process characteristics and the individuality of production devices should be considered when process data is analyzed. Since such changes and individuality are expressed as the difference of the correlation among variables, samples should be clustered on the basis of the correlation among variables. In this work, a new sample clustering method is proposed on the basis of spectral clustering. Spectral clustering is a graph partitioning method and it can be used for sample classification when an affinity matrix of a weighted graph is given. The proposed method can cluster samples based on the correlation among variables without teacher signals. The usefulness of the proposed method is demonstrated through a numerical example and a case study.