Recursive Gaussian Mixture Models for Adaptive Process Monitoring

Gaussian mixture models (GMM) have recently been introduced and widely used for process monitoring. This paper intends to develop a new recursive GMM model for adaptive monitoring of processes under time-varying conditions. Two model updating schemes with/without forgetting factors are both proposed. Bayesian inference probability index is used as the monitoring statistic in both of the continuous and batch process monitoring models. In order to reduce the online computational complexity, an updating strategy for both determinant and inverse of the covariance matrix during the monitoring process is particularly formulated. According to the simulation results of two case studies, efficiencies of both recursive modeling and adaptive monitoring performances are evaluated.