PATTERN RECOGNITION AND MULTIVARIATE STATISTICS TO CONTROL AND MONITOR PRODUCTION PLANTS
Publication Date: 2017-Aug-08
The IP.com Prior Art Database
Most production process, whether they are Air Separation Units or Steam Methane Reformers, are nonlinear in nature, i.e., a change in input variables (air flow, for example) may not always produce the same response in the associated output variables (GOX purity, for example). Currently, while implementing Advanced Process Control (APC) technology in plants, this nonlinearity is neglected due to technology limitations and a linear model is used to control the plant. Utilization of a linear model enables a simple implementation of the APC technology which can be tweaked over time to fit the plant's needs. However, use of a linear model based on a single operating target limits its use in optimizing the process at different production targets (called as steady state operation) if needed. In addition, custom programming is used to move the plant from one production target to another (called as ramping operation). This hybrid approach combines clustering and statistical techniques: 1) clustering techniques segment the nonlinear process into linear regions and 2) the linear regions are then analyzed using techniques such as Principal Component Analysis (PCA) to identify the key variables which describe the process. The techniques are appropriate for online systems and real-time analysis for plant KPI monitoring, sensor fault detection, and trip abatement.
Nothing specific to the application mention...