Using estimability analysis to rank and select model parameters

Mathematical models can be used to support the development and optimization of chemical processes. However, modelers often face challenges in estimating all the model parameters due to identifiability and estimability issues. This is typically the case of complex and/or over-parameterized models, resulting in some parameters being unidentifiable or difficult to be precisely estimated. Moreover, data are usually limited, while running additional experiments might be expensive or impractical. To tackle this issue, modelers often estimate only a subset of all the model parameters. In this work, a systematic workflow for parameter estimability proposed in the literature (Wu et al., 2011, International Journal of Advanced Mechatronic Systems, 3, pp. 188-197) is implemented and critically evaluated. The methodology is applied to two case studies of industrial interest: (i) a fermentation process, and (ii) a process for the production of urethane. Different levels of uncertainty on parameter values and experimental noise are taken into account. A robustness test is used to assess the sensitivity of the methodology to initial parameter guesses. Results suggests that parameters ranking and subset selection depends significantly on both the initial guesses of the model parameters and sensors noise, and that repeating the methodology can improve the results and the precision of parameter estimates.