Qualitative interpretation and compression of process data using clustering method

Abstract

This paper presents a new qualitative data interpretation and data compression method, which is based on modified adaptive k-means clustering algorithm. Conventional qualitative data interpretation methods that are based on control charts, such as Shewart, CUSUM, and EWMA control charts, are focused upon detection of changes from steady state values, so they are not suitable for describing transient or dynamic behavior. But the proposed method continuously updates its detection limit, or center of cluster, so it can handle transient or dynamic behavior. Thus it can be applied to fault diagnosis of chemical process by combining with cause-effect digraph model, RCED(Reduced Cause Effect Digraph). The usefulness of the proposed data interpretation method and cause-effect digraph model are illustrated using their application to the water supply unit of a utility boiler plant. The proposed data interpretation method can be used for not only changed detection but also data compression. As the proposed method can store the value that minimizes the SSE between the retrieved data and the original data, it shows better data compression result compared to other conventional data compression methods such as Box Car, Backward Slope, Combined Box Car and Backward Slope algorithm and Swinging Door Trending.