Hybrid Intelligence: The Fusion of Science-Based and Machine Learning Models
Authors:
- Eleni LavasaELEleni Lavasa
- Theodora ChrysoulaTCTheodora Chrysoula
- Vasileios GkolemisVGVasileios Gkolemis
- Charalambos SinnisCSCharalambos Sinnis
- Theodore DalamagasTDTheodore Dalamagas
- Sotiris KoussourisSKSotiris Koussouris
- Nefeli BountouniNBNefeli Bountouni
- Konstantinos PerakisKPKonstantinos Perakis
- Viktor DaropoulosVDViktor Daropoulos
- Stratos KeranidisSKStratos Keranidis
- Charalambos LambriCLCharalambos Lambri
- Giorgos IoannouGIGiorgos Ioannou
Open Access
- 7 March 2026
- book chapter
- Published by Springer Nature
- p. 137-173
Abstract
Process models and parameter estimation have long been fundamental tools in domains such as manufacturing, robotics, power plants, and biotechnology. Early modeling approaches relied on complex systems of mathematical differential and algebraic equations, known as science-based models, to capture process knowledge and scientific expertise. These models leverage physical and chemical properties, static and dynamic behaviors, and causal relationships among observed quantities to support predictive control and operational optimization. As white-box models, they provide transparency by uncovering the inner logic and decision-making steps of the process. However, the advent of Industry 4.0 has brought a surge in available data from industrial processes, driving the rapid growth of machine learning (ML) models. Those models excel at discovering patterns and nonlinear relationships in data, but often they are black-box models, i.e., they lack interpretability. To combine the strengths of science-based and ML models, hybrid models have emerged as a powerful solution. By integrating the transparency and domain knowledge of science-based approaches with the adaptability and predictive capabilities of ML, hybrid models enhance accuracy, robustness, and scalability. This chapter explores the foundations of hybrid models, their development, and applications, providing a comprehensive perspective on their transformative potential across various scientific and engineering domains. This work is done in the context of the AI-DAPT EU project.Keywords
Funding Information
- University of Galway
This publication has 95 references indexed in Scilit:
- Hybrid model with optimal features for non-invasive blood glucose monitoring from breath biomarkersBiomedical Signal Processing and Control, 2023
- Scientific Machine Learning Through Physics–Informed Neural Networks: Where we are and What’s NextJournal of Scientific Computing, 2022
- Physics informed neural networks for control oriented thermal modeling of buildingsApplied Energy, 2022
- A hybrid science‐guided machine learning approach for modeling chemical processes: A reviewAIChE Journal, 2022
- Physics-informed linear regression is competitive with two Machine Learning methods in residential building MPCApplied Energy, 2022
- Surrogate-based aerodynamic shape optimization for delaying airfoil dynamic stall using Kriging regression and infill criteriaAerospace Science and Technology, 2021
- Machine Learning: Algorithms, Real-World Applications and Research DirectionsSN Computer Science, 2021
- Approximate Bayesian computational methodsStatistics and Computing, 2011
- Bayesian Learning for Neural Networks1996
- Numerical heat transfer and fluid flow. Von S. V. Patankar. Hemisphere Publishing Corporation, Washington – New York – London. McGraw Hill Book Company, New York 1980. 1. Aufl., 197 S., 76 Abb., geb., DM 71,90Chemie Ingenieur Technik, 1981