Federico Galvanin


Development and Application of Machine Learning‐Assisted Techniques for Kinetic Model Identification
The quantitative description of the behaviour of reacting systems requires the identification of an appropriate set of kinetic model equations. This identification problem poses substantial challenges to the modeller as there may be a large number of potentially appropriate kinetic model structures to consider. Model-based design of experiments (MBDoE) procedures for model discrimination [1,2] can be used for minimising the time and effort required in the experimentation to quickly identify a suitable model among a set of candidates. However, MBDoE techniques have two main limitations: i) candidate kinetic models may be affected by identifiability issues [3], so that the estimation of their parameters for model discrimination purposes may be impractical or even impossible; ii) the efficiency of these techniques is strictly related to the model reliability in the experimental space of intended model utilisation.  

In this seminar we will see how machine learning techniques and MBDoE techniques can be coupled to overcome these issues in an integrated framework for rapid kinetic model identification. In the framework, new machine learning-assisted model identification techniques recently developed at UCL and based on Artificial Neural Network (ANN) classifiers are used for recognising appropriate kinetic model structures from limited set of data [4]. These methods do not require the fitting of kinetic parameters and they are well suited when a large number of candidate kinetic mechanisms needs to be tested. Model-based data mining techniques based on support vector classifiers [5] are used to detect outliers in the data and build “reliability maps” to screen out regions in the experimental design space where candidate kinetic models are weak and not suitable for MBDoE application. In the seminar these techniques will be illustrated and discussed through case studies from chemical kinetics with the specific aim to bridge the gap between data-driven modelling and mechanistic kinetic modelling using machine learning. 


  1. Schwaab, M., Silva, F.M., Queipo, C.A., Barreto Jr., A.G., Pinto J.C. (2006). A new approach for sequential experimental design for model discrimination. Chem. Eng. Sci., 61, 5791.
  2. Kremling, A. et al. (2004). A benchmark for methods in reverse engineering and model discrimination: problem formulation and solutions. Genome Res., 14, 1773. 
  3. Walter, E, Pronzato, L. (1997). Identification of Parametric Models from Experimental Data. Communications and Control Engineering Series. London (UK), Springer. 
  4. Quaglio M., Roberts L., Jaapar M. S., Dua V., Fraga, E. S., Galvanin F. (2020). An artificial neural network approach to recognise kinetic models from experimental data. Comp. Chem. Eng., 135, 106759.  
  5. Quaglio M., Fraga E. S., Cao E., Gavriilidis A., Galvanin F. (2018). A model-based data mining approach for determining the domain of validity of approximated models, Chemometrics and Intelligent Laboratory Systems 172, 58.


Dept of Chemical Engineering, University College London

Dr. Federico Galvanin is a Lecturer in Chemical Engineering at UCL. He obtained his Master’s degree in Chemical Engineering in 2006 the PhD degree in Industrial Engineering in 2010 from the University of Padova (Italy). His research interests lie at the interface between mathematical modelling and experimentation, with a specific expertise in the development and application of computational methods for fast model development using physics-based modelling, statistical planning and machine learning techniques. He has published >50 articles in peer-reviewed journals and several book chapters on theory and applications of model-based design of experiments (MBDoE) techniques. He is a member of the Centre for Process Systems Engineering (CPSE) and affiliate of the Institution of Chemical Engineers. In 2014 he was awarded the 1st prize in “Process Systems Enterprise Ltd. Model-Based Innovation” for the most advanced use of equation-oriented simulation in process systems engineering applications.

Comments are closed.