Scale-Space Filtering for Qualitative Interpretation of Real-Time Process Data

Abstract This paper describes a qualitative interpretation method, which is used for extracting qualitative information from numeric sensor data. Firstly, whether any change has occurred in chemical process data is determined by using the CUSUM (Cumulative SUMmation) test. From the sign of the first derivatives of the process variables, sensor patterns can be classified into the seven primitives. Secondly, extraction of the trends of the data employing the modified scale-space filtering is performed. The recursive fonn reduces the calculation cost of the real-time scale-space filtering and solves the endpoint problem. The proposed method was tested for the artificial patterns and the simulated data of a evaporator process.