Identification of the isotherm function in chromatography using CMA-ES

Mohamed Jebalia, Anne Auger, Marc Schoenauer, François James, Marie Postel

Abstract

This paper deals with the identification of the flux for a system of conservation laws in the specific example of analytic chromatography. The fundamental equations of chromatographic process are highly non linear. The state-of-the-art Evolution Strategy, CMA-ES (the Covariance Matrix Adaptation Evolution Strategy), is used to identify the parameters of the so-called isotherm function. The approach was validated on different configurations of simulated data using either one, two or three components mixtures. CMA-ES is then applied to real data cases and its results are compared to those of a gradient-based strategy.

Introduction

In chemical engineering, chromatography is a cornerstone technique for separating complex mixtures into pure components. The efficiency of this process hinges on understanding the "isotherm function"—a mathematical description of how different chemicals interact with the stationary phase inside a chromatography column. Identifying this function from experimental data is a notoriously difficult inverse problem. Traditional gradient-based optimization algorithms often fail because the "fitness landscape" is non-convex and multi-modal, leading them to get trapped in local optima.

Methodology: Optimization via CMA-ES

The authors propose using the Covariance Matrix Adaptation Evolution Strategy (CMA-ES), a state-of-the-art stochastic optimization algorithm designed for continuous domains. To evaluate a candidate isotherm, the authors use the Godunov scheme to numerically solve the underlying non-linear Partial Differential Equations (PDEs) and generate a simulated chromatogram.

A key innovation in this study is the use of a CMA-ES restart strategy. If the algorithm converges to a point that is not sufficiently optimal, it restarts with an increased population size. This allows the search to escape local minima and explore the parameter space more effectively.

Convergence of the CMA-ES algorithm across multiple restarts, demonstrating the ability to reach high-precision fitness values near 10^-14.

Results and Discussion

The approach was evaluated using both simulated and real experimental data to test its robustness and precision.

Performance on Simulated Data

On artificial problems involving 1, 2, and 3-component mixtures, CMA-ES reached fitness values as low as $10^{-15}$, effectively recovering the exact parameters used to generate the data. This level of precision highlights the algorithm's capability in controlled environments.

Application to Real Experimental Data

CMA-ES was applied to a ternary mixture of benzylalcohol (BA), 2-phenylethanol (PE), and 2-methylbenzylalcohol (MBA). In this 6-parameter problem, CMA-ES achieved a fitness value of $8.32 \times 10^{-3}$, significantly outperforming the best gradient-based result of $10.7 \times 10^{-3}$. While gradient methods often required researchers to manually fix certain parameters to achieve convergence, CMA-ES handled all parameters simultaneously with ease.

Comparison between experimental data (MBA exp) and the identified chromatogram (MBA ident) for a ternary mixture component.

Conclusion

CMA-ES proves to be a highly robust and effective tool for identifying the non-linear parameters of isotherm functions in chromatography. Unlike traditional gradient-based methods, it reliably finds global optima in multi-modal search spaces and handles complex mixtures without getting stuck in local minima. While computationally more expensive than calculus-based methods, its reliability and invariance properties make it a superior choice for complex chemical engineering inverse problems. Future work may explore multi-objective optimization to simultaneously minimize errors on both chromatogram shapes and direct isotherm measurements.

Reference: Jebalia, M., et al. (2007). Identification of the Isotherm Function in Chromatography Using CMA-ES. arXiv:0710.0322v1 [math.OC].

Summary updated

BibTeX
@inproceedings{jebalia2007identification,
  author = {Mohamed Jebalia and Anne Auger and Marc Schoenauer and François James and Marie Postel},
  title = {{Identification of the isotherm function in chromatography using CMA-ES}},
  booktitle = {IEEE CEC},
  year = {2007},
  doi = {10.1109/cec.2007.4425031},
}
Figure