(546e) Fault Detection for Pressure Swing Adsorption Processes

Authors: 
Lee, J., Auburn University
Kumar, A., Praxair, Inc.
Wang, J., Auburn University
He, Q. P., Auburn University
Fault Detection for Pressure Swing Adsorption Processes

Jangwon Lee, Ankur Kumar, Jesus Flores-Cerrillo, Jin Wang, Q. Peter He

Pressure swing adsorption (PSA) is widely used to separate a product from impurities in a variety of different fields. Due to the complexity of PSA operations, faults can occur at different parts and/or different steps of the process. Thus, efficient fault detection is a critical component for ensuring effective and safe operation of the process. One of the key characteristics of PSA process is that it is operated in a periodic or cyclical fashion, making fault detection significantly more challenging than that of processes operated at steady-state. Although periodic operations have been used widely in chemical and petrochemical industries, the monitoring of these operations has received little attention compared to continuous or batch processes. Hence, approaches targeting the fault detection in PSA process are worth further research and development [1].

In this study, we propose a statistics pattern analysis (SPA) based statistical process monitoring (SPM) for PSA processes, which is based on the SPA process modeling and monitoring framework [2, 3]. The main difference between the mainstream SPM approaches, such as those based on principal component analysis (PCA), and SPA based fault detection method is that the traditional SPM approaches monitor process variables while SPA monitors the statistics of process variables. In other words, for the traditional SPM approaches, fault is detected by examining the variance−covariance of the process variables; while for SPA, fault is detected by examining the variance−covariance of the statistics of the process variables (e.g., mean, standard deviation, coefficient of variation, etc.). By using various statistics including high order statistics, SPA model can monitor process nonlinearity which makes SPA method more effective than the traditional linear SPM methods.

In this work, we investigate the capabilities and limitations of SPA based fault detection method for monitoring PSA processes. We show that the SPA based method can address some challenges in monitoring periodic process, such as eliminating many data preprocessing steps including trajectory alignment or synchronization, etc. Different strategies for dividing PSA cycles into steps and extracting various statistics from each step are studied and their impacts on fault detection performance are compared and discussed. Finally, different strategies for reducing false alarms are investigated and their performances are compared.

Reference:

[1] R. Wang, T. F. Edgar and M. Baldea, “A Geometric Framework for Monitoring and Fault Detection for Periodic Processes”, AIChE J., vol. 63, pp.2719-2730, 2017.

[2] J. Wang and Q. P. He, “Multivariate process monitoring based on statistics pattern analysis,” Ind. Eng. Chem. Res., vol. 49, pp. 7858–7869, 2010.

[3] Q. P. He and J. Wang, “Statistics Pattern Analysis - A New Process Monitoring Framework and Its Application to Semiconductor Batch Processes,” AIChE J., vol. 57, pp. 107–121, 2011.