A quality spectra-based SVDD method for multi-dimensional quality fluctuation evaluation in complex industrial process

Abstract

To address the problem that fluctuation caused by multi-dimensional quality-related process variables is difficult to evaluate, this paper proposes a quality spectra-based support vector data description (SVDD) method for multi-dimensional quality fluctuation evaluation in complex industrial process. Firstly, based on quality spectra modelling, a multi-dimensional state feature vector of quality fluctuation is constructed from the perspective of quality spectra and statistical distribution, which is used to characterize the difference between normal and abnormal fluctuation. Then the quality fluctuation evaluation model is established based on SVDD algorithm. On this basis, the spherical distance between the sample to be evaluated and the quality evaluation model is calculated to measure the quality fluctuation degree. Finally, a case for TE process data sets is used to verify the quality fluctuation evaluation approach, it reveals that the proposed method could evaluate the degree of quality fluctuation, which could provide some theoretical guides for quality stability control.