Enhancing data-driven fault detection through extended attribute variables

Due to the high demand for safety and cost efficiency, process monitoring has been well studied. One of the most popular approaches for process monitoring is data-driven fault detection, which usually do not use process knowledge. This paper presents a preprocessing method to combine process knowledge with data-driven fault detection of chemical plant. The method provides a rule to generate extended attribute variables, and the better fault detection is expected with the extended dataset by usual data-driven approach such as a PCA based method. The method was successfully applied to fault detection of the Tennessee Eastman plant simulation benchmark problem.