Advanced error modeling and Bayesian uncertainty quantification in mechanistic liquid chromatography modeling.

Area of Science:

• Chemical Engineering
• Process Modeling
• Chromatography

Background:

• Mechanistic chromatography process models often overlook experimental errors beyond detector noise.
• Uncertainty in model parameters and predictions can arise from factors like pump delays and variable feed composition.
• Accurate uncertainty quantification is crucial for reliable process simulation and optimization.

Purpose of the Study:

• To develop and present an uncertainty quantification method for mechanistic chromatography process models.
• To incorporate realistic experimental errors, such as pump delays and variable feed composition, into model calibration.
• To determine the probability distribution of calibrated model parameters and their impact on predicted chromatograms.

Main Methods:

• Bayes' theorem and Markov chain Monte Carlo (MCMC) with an ensemble sampler were employed.
• The method quantifies uncertainty by determining the probability distribution of model parameters.
• The approach accounts for multiple realistic sources of experimental error.

Main Results:

• The developed uncertainty quantification method demonstrates robustness and extensibility.
• The method was validated using both synthetic and industrial chromatography data.
• The probability distributions of model parameters and their impact on predictions were successfully determined.

Conclusions:

• The presented method effectively addresses the limitations of current chromatography modeling regarding experimental error.
• This approach enhances the reliability of mechanistic models by providing uncertainty estimates for parameters and predictions.
• The open-source software implementation facilitates broader adoption and application in chromatography process development.