Abstract:
This paper presents an approach which uses the interval type 2 fuzzy C means cluster to classify the process data, and then, adaptively selects the cluster number, so that different models can be distinguished. The local tangent space alignment algorithm (LTSA) is adopted to reduce the dimensions of the data of each cluster, which is further utilized to build multi model via the support vector data description algorithm (SVDD), and obtain corresponding statistical magnitude and statistical limit. Thus, the offline modeling is achieved. During the online monitoring, the first is to judge which model the process data belong to, and then, to judge whether it is a fault data by calculating its statistical magnitude. Finally, the simulation on the process data of ethylene cracking furnace is made to verify the feasibility of the proposed approach in this work.