Thesis
Robust sampling time design for uncertain dynamic systems
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2026
- Thesis identifier
- T18183
- Person Identifier (Local)
- 201767789
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- Data-based modelling of chemical and biochemical systems, where the underlying physical mechanisms are generally well understood, relies on accurate parameter estimation from experimental data. However, parameter estimation for nonlinear dynamic models is often ill-conditioned due to model nonlinearity, parameter correlations, and limited experimental resources. Since model prediction reliability depends critically on data quality, designing informative experiments is essential for accurate model calibration. Optimal experimental design (OED) provides a systematic framework for selecting experimental conditions that maximise data information content for objectives such as identifiability analysis, parameter estimation, and model validation. However, classical OED relies on a nominal model and is therefore sensitive to model uncertainties and process variations, motivating the development of robust experimental design (RED) methods. This thesis develops robust sampling time design strategies for parameter estimation in nonlinear dynamic systems under parametric and input uncertainties. Challenges associated with large parametric uncertainty and input uncertainty are addressed within an optimisation based design framework. The main contributions are summarised as follows: • Developed robust sampling time design methods based on maximin and pseudo- Bayesian criteria for systems with parametric uncertainty. The corresponding problem formulations, numerical solution strategies, and implementation procedures are presented. • Developed a distribution-based robustness evaluation framework for sampling time design, enabling consistent quantitative assessment using multiple statistical performance metrics. This framework is applied to compare robust experimental design methods using two nonlinear dynamic systems: an enzyme reaction network and an enzymatic biodiesel production process. • Proposed a midterm sampling time redesign strategy to overcome key limitations of existing robust design methods. The experimental horizon is divided into sequential zones, within which OED or RED is performed using updated parameter estimates from previously collected data. This enables adaptive experimental design within a single experimental run without repeated trials. Three redesign strategies are developed: equally spaced redesign, auto-updating redesign based on practical identifiability analysis, and condition-triggered redesign activated by predefined operational changes. • Developed a sampling time design framework based on the misspecified Cramér– Rao bound (MCRB) to address input uncertainty. The proposed framework accounts for changes in pseudo-data dispersion, departures from the assumed measurement distribution, and temporal correlations induced by nonlinear propagation of input uncertainty. A full trajectory (stacked-measurement) MCRB criterion is formulated for sampling time selection. Comparisons with classical FIM-based OED, inputuncertainty- specific methods, and standard RED formulations show that the MCRBbased criterion provides a more appropriate treatment of distributional misspecification under input uncertainty. This thesis establishes a coherent framework for robust sampling time design for nonlinear dynamic systems under model and input uncertainties. By integrating robust design principles, adaptive experimental strategies, and performance bounds under data-model misspecification, these contributions improve the reliability and practical applicability of MBDoE and provide a foundation for more efficient experimental design in complex systems.
- Advisor / supervisor
- Yue, Hong
- Resource Type
- DOI
- Funder
Relations
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
PDF of thesis T18183 | 2026-09-30 | Public | Download |