A Multi-Model MLLE-PCA Method for Unstable Industrial Process Monitoring

Abstract

It is quite a challenge to monitor and diagnose an unstable industrial process, because the changes of the industrial process bring the local structure changes of process data. Traditional process monitoring methods train and model the process as a whole. Such a model shows the overall structure of the process data, but ignore the local characteristics. In order to construct the local characteristics of the unstable process, Modified Locally Linear Embedding (MLLE) is introduced into the PCA process monitoring to reveal the local data structures. At the same time, to solve the error mapping problem of anomaly points, this paper introduces a multi-model framework and constructs a new Multi-Model MLLE-PCA method for unstable industrial process monitoring. Compared with the traditional method, the proposed method performances better in simulation.