(150c) Semi-Rigorous Models for Virtual Analyzers in Distillation Units

Model predictive multivariable control typically require proper measure for stream quality, either in terms of product composition or distillation curve point. Different technologies for online analysers are available and can provide quality measure with sufficient sampling frequency to be directly included in model predictive controllers. Nonetheless, analyser is generally expensive equipment and frequently not available on some important process streams, which relies on low frequency laboratory sampling. In such cases, several different methodologies for building inferential/virtual analyser have been proposed and used in the chemical industry, and particularly applied to the distillation processes, basing on available instrumentation. This paper describes semi-rigorous methodology applied to simple distillation column and atmospheric fractionator (as example, in refining industry, crude units, main fractionator after FCC reactor, etc).

The starting point is availability of a digital twin of the unit. From the digital twin, we can extract chemical and physical parameters necessary to build a set of simplified core equations for the virtual analyser. The simplified equation set is built in with some important characteristics:

• Can be solved only with the available process measurements (temperatures, pressures, flows, etc.)
• Can be solved high frequency, and with no risk of not converging to a solution, i.e. continuous and reliable predictions
• When running online, can provide a simple mechanism for reconciliation and feedback from available quality measures (e.g. lab sampling)

Key benefit from the methodology it that, being based on digital twin, the virtual analyser is relatively quick to be deployed and does not require big training datasets for initial data reconciliation. The paper will include some implemented case studies and their results.