BioLaMer project team members Monesh Thirugnanasambandam and Roshanak Agharafeie from NOVAID FCT, Lisbon, represented the hybrid models developed within the project at the MADEAI Conference on Modelling, Data Analytics, and AI in Engineering, held in Porto, Portugal, from July 2-5, 2024.
Conference Briefing:
MadeAI conference is a premier event that brought together global leaders and researchers from computer science, engineering, and mathematics to explore the intersection of modelling, data analytics, and AI in engineering. This conference provided a unique platform to delve into the fundamental principles and methodologies of these topics, investigate cutting-edge applications across various engineering domains, and foster interdisciplinary collaboration. Attendees discovered innovative solutions and new opportunities in aerospace, automotive, chemical, civil, electrical, mechanical engineering, and other fields, as they collaborated to define the future of technology and problem resolution.
During this prestigious event, Monesh Thirugnanasambandam delivered an insightful oral presentation on “A Comparative Study of Physics-Informed Neural Networks and Traditional Semiparametric Hybrid Modeling.” His session attracted considerable attention and sparked engaging discussions among the attendees. Here are the key highlights from his presentation:
- A case study for a microbial logistic growth model in the stirred tank bioreactor simulation employing two different modeling approaches: Traditional hybrid semiparametric and Physics-informed neural network.
- In this study, a comparison of predictive accuracy between the two models revealed noteworthy findings, emphasizing the advantages of traditional hybrid semiparametric models in this specific bioreactor simulation.
- Future work will focus on tackling complex problems in bioreactors and addressing the intricacies of microbial growth dynamics and bioprocess optimization using these approaches.
Roshanak Agharafeie delivered an informative oral presentation on the title “Deep Neural Network Modelling in Supercritical Co2 Extraction Process”. Her talk highlighted the significant findings from our model developed for the scCO2 extraction process. Here is the summary of the model developed and results:
This extraction method is a safe and environmentally friendly method for biomanufacturing, avoiding chemical solvents, and effectively extracting heat-sensitive substances. However, challenges arise in accurately defining the relationships between mass transfer coefficients, flow conditions, properties of scCO2, and physiochemical properties of the extracted substance. To address these limitations, our model integrates a deep feedforward neural network with intraparticle and macroscopic material balance equations, represented as a system of Partial Differential Equations (PDEs).
Key results shared at the event;
- The HNN model predicts the yield of the extraction with less than 10% error.
- It provides insights into the extraction rate at different levels inside the extraction column.
- It offers good optimization capabilities.
