Skip to main content
— Presentation at PSE Asia 2016 —
September 6, 2016
The 7th International Symposium on Design, Operation and Control of Chemical Processes (PSE Asia 2016) was held in The University of Tokyo, Japan, from July 24 to 27, 2016. PSE Asia has been held biennially for researchers mainly from Japan and Asia to discuss about chemical process control. In PSE Asia 2016, there were 134 oral presentations and 36 poster presentations.
In PSE Asia 2016, Hitachi R&D group made total 5 oral presentations in collaboration with Universiti Teknologi PETRONAS to discuss anomaly detection and cause analysis technologies in oil & refinery plants. One of these presentations, I made a presentation tilted "Cause Analysis of Representative Troubles at Distillation Tower Using Linear Regression" to discuss cause analysis on crude distillation unit (CDU) in oil & refinery plants.
Fig. 1 Cause Analysis on Tray upset
One of issues in Oil & Refinery plants is lowering operating efficiency of CDU due to off-spec products. The off-spec products can be produced by contamination of the petroleum products, which occurs due to anomalies in CDU such as tray upset. Tray upset is a phenomenon that product in a tray overflows, and causes contamination of the product at other tray located below the tray. In order to improve the efficiency, a facility diagnosis method to detect cause of CDU anomalies was proposed. Although diagnosis methods have been studied to detect the anomalies, they can result in false negatives due to being insensitive to small process upsets.
Fig. 2 Pilot equipment at Universiti Teknologi PETRONAS
We proposed a facility diagnosis method based on linear regression to detect cause of small upsets, such as tray upset (Figure 1). The method requires abstract models which describe physical phenomena in CDU but does not require learning data. The proposed method detects facility anomalies and causes by observing the phenomena change with the models.
Binary distillation experiments using a mixture of ethanol and water were carried out on pilot equipment at Universiti Teknologi PETRONAS confirmed the method could detect causes of small upset anomalies such as tray upset with below 10% of average false positive ratio (Figure 2).
The research is under collaboration research with Universiti Teknologi PETRONAS. We would like to express our appreciation to the assistance and support by Universiti Teknologi PETRONAS.