Scale-Up of Chemical Process Models: Re-Calibrating Fundamental Models using Data Mining

A.J. Owens (USA)


Chemical Process Modeling; Scale-Up; Support Vector Machines


To understanding a chemical kinetic reaction scheme, a large number of experiments are often performed in a ‘Lab’ or small-scale reactor. The experimental results can be represented by a fundamental mathematical model, typically a series of ordinary differential equations (ODEs) that are solved numerically. When the real process is ‘scaled up’ to a much larger ‘Plant’ reactor, some of the underlying model’s process parameters can change to new, unknown values. This paper proposes a simple method to re-calibrate the fundamental model based on lab data to accurately represent the plant process. The method involves running the lab scale fundamental model at a large number of design points, recording the simulation predictions and derivatives with respect to the model inputs. Only a few experiments need to be run in the plant, and those data are combined with the simulations in a specially trained support vector machine (SVM) that simultaneously adjusts its parameters to fit both sets of data. The resulting empirical model gives an accurate representation of the plant scale process.

Important Links:

Go Back