Abstract
Root cause analysis is an important step in process monitoring for the disclosure of causes of process disturbances. In the past research and practice, transfer entropy has been widely adopted for its capability to handle process nonlinearity. However, the conventional transfer entropy is not robust to noise and computational inefficient, which affects its application performance. These problems can be solved by incorporating the technique of symbolization. To the best of our knowledge, symbolic transfer entropy (STE) has seldom been implemented in the field of process monitoring. In this work, the concepts of statistical process control (SPC) and STE are integrated to achieve satisfactory diagnosis results. As known, SPC charts, e.g. Shewhart charts, exponential weighted moving average charts, and cumulative sum charts, are useful tools for process monitoring, where the sample locations on the control chart indicate certain process status. Therefore, it is reasonable to symbolize process measurements according to such information. After symbolization, STE is conducted to reveal the causality among process variables which is then visualized with a signed directed graph. The case study on the Tennessee Eastman process shows that the control chart-based STE possesses a capability to understand the causes of disturbances and offers the convenience of root cause analysis.