This work is motivated by challenges in data-based modelling of complex systems due to limited information of sparse and noisy experimental data. Optimal experimental design (OED) techniques, which aim at devising necessary experiments to generate informative measurement data to facilitate model identification, have been investigated comprehensively.The limitations of existing experimental design approaches have been extensively discussed, based on which advanced experimental design methods and efficient numerical strategies have been developed for improved solutions.Two case study biochemical systems have been used through the research investigation, one is an enzyme reaction system, the other one is a lab-scale enzymatic biodiesel production system. The main contributions of this PhD work can be summarised as follows:Single objective experimental designs by considering one type of design factors, i.e. input intensity, measurement set selection, sampling profile design, respectively, has been formulated and numerical strategies to solve these optimisation problems have been described in detail. Implementations of these design methods to biochemical systems have demonstrated its efficiency in reducing parameter estimation errors.A new OED strategy has been proposed to cope with OED problems including multiple design factors in one optimisation framework. An iterative two-layer design structure is developed. In the lower layer for observation design, the sampling profile and the measurement set selection are combined and formulated as a single integrated observation design problem, which is relaxed to a convex optimization problem that can be solved with a local method.Thus the measurement set selection and the sampling profile can be determined simultaneously. In the upper layer for input design, the optimisation of input intensities is obtained through stochastic global searching. In this way, the multi-factor optimisation problem is solved through the integration of a stochastic method, for the upper layer, and a deterministic method, for the lower layer.A new enzyme reaction model has been established which represents a typical class of enzymatic kinetically controlled synthesis process. This model contains important kinetic reaction features, moderate complexity, and complete model information. It can be used as a benchmark problem for development and comparison of OED algorithms. Systematic analysis has been performed in order to examine the system behaviours, and the dependence on model parameters, initial operation conditions.Structural identifiability and practical identifiability of this system have been analysed and identifiable parameters determined. The design of experiment for the enzyme reactionsystem by considering different types of design variables have been investigated. The parameter estimation precision can be improved significantly by using the proposed OED techniques, compared to the non-designed condition.The OED techniques are numerically investigated based on a lab-scale biodiesel production process with real experimental data through research collaboration with DTU in Denmark. The OED applications on this real system model allow to examine the effectiveness and efficiency of those new proposed OED methods. The measurement set selection and the sampling design of this system are developed which provide detailed instructions on how to improve experiments through OED.Also, the sensitivity analysis and parameter identifiability analysis are conducted; and their impacts to experimental design are clearly identified.
|Date of Award||6 Aug 2018|
- University Of Strathclyde
|Supervisor||Hong Yue (Supervisor) & Peter Halling (Supervisor)|