Fault Detection in Pressure Swing Adsorption Systems

Abstract

Over the years, there has been a consistent increase in the amount of data collected by systems and processes in many different industries and fields. On the other hand, there is a growing push towards revealing and exploiting of the collected data. The chemical processes industry is one such field, with high volume and high-dimensional time series data. The massive amount of data can be used for better control and production. Because of the high level of complexity in chemical processes, mathematical modeling is mostly unachievable and impracticable. A number of model base on process data have been suggested and developed by researchers for batch and continuous processes. Compared to these type of processes, cyclic processes have not received enough attention in the literature and still, it is a relatively intact area for developers in fault detection and process monitoring. Moreover, because of the different nature of the batch and continuous, well-developed monitoring methods and frameworks in those processes cannot be employed to the periodic processes. Therefore, a new multivariate method based on combined Principal Component Analysis and Statistical Pattern Analysis framework is proposed to help to fill the existing gap in the monitoring of cyclic processes and to overcome fault detection issues.