Welcome to the ENBIS Spring Meeting
Quality by Design (QbD) & Process Analytical Technology (PAT): Statistical, AI, and “Grey” Approaches
Coimbra, Portugal, May 29-30, 2025
Aim of the Spring Meeting
Quality by Design (QbD) plays an important role in the modern industry. Often, it goes hand in hand with technological solutions that secure the supervision and stability of the processes in the design space (DS), such as spectroscopy, namely near-infrared, Raman, fluorescence, or UV, collectively called Process Analytical Technology (PAT). QbD and PAT initiatives have benefited from concepts, methodologies and tools arising from different corners of the data-driven sciences, such as statistics, chemometrics, and machine learning. The integration of first principles and existing knowledge with empirical modelling, called grey modelling, is another area of current active research that may bring important contributions to QbD/PAT. Inevitably, Artificial Intelligence will also play a role in the future of these fields, but which exactly is an open question. Therefore, it is both important and opportune to assess how these different approaches can further contribute to improving or reinventing QbD and PAT initiatives, either isolated or cooperatively. The main goal of the ENBIS Spring Meeting 2025 is to foster high-level discussions on these and other related topics.
All stakeholders, from graduate students to professionals, from researchers to industrial managers, are invited to participate actively. We welcome contributions in the following areas (the list is not exhaustive), with a focus on their application in industry in the scope of QbD/PAT:
- Design of experiments (physical systems or in silico)
- Active learning, Bayesian optimization and derivative-free optimization
- Design spaces under uncertainty (Bayesian and frequentist approaches)
- Explainable and Generative AI
- Transfer learning
- New grey (hybrid) modelling architectures
- Digital Twins
- Physics-informed neural networks (PINNs)
- Chemometrics: classical and new AI methods
- Integrated approaches of the above
Contact information
For any question about the meeting venue and scientific programme, registration and paper submission, feel free to contact the ENBIS Permanent Office : office@enbis.org.
Local Organizing Committee:
- Marco S. Reis (Chair)
- Tiago J. Rato
- João Coutinho
- Rúben Gariso
Programme Committee:
- Marco S. Reis (Chair), University of Coimbra, Portugal
- Raffaele Vitale (Co-Chair), Université de Lille, France
- Jacqueline Asscher, Kinneret College, Israel
- Alberto Ferrer, Universidad Politecnica de Valencia, Spain
- Pierantonio Facco, University of Padova, Italy
- Valeria Fonseca, Software Competence Center Hagenberg, Austria
- Sonja Kuhnt, Dortmund University of Applied Sciences and Arts, Germany
ENBIS Spring Meeting 2025 Highlights
Plenary speakers
Antonio Del Rio Chanona (Imperial College, UK)
LLM and human-in-the-loop Bayesian optimization for chemical experiments
Bayesian optimization has proven effective for optimizing expensive-to-evaluate functions in Chemical Engineering. However, valuable physical insights from domain experts are often overlooked. This article introduces a collaborative Bayesian optimization approach that integrates both human expertise and large language models (LLMs) into the data-driven decision-making process. By combining high-throughput Bayesian optimization with discrete decision theory, experts and LLMs collaboratively influence the selection of experiments via a human-LLM-in-the-loop discrete choice mechanism. We propose a multi-objective approach to generate a diverse set of high-utility and distinct solutions, from which the expert, supported by an LLM, selects the preferred solution for evaluation at each iteration. The LLM assists in interpreting complex model outputs, suggesting experimental strategies, and mitigating cognitive biases, thereby augmenting human decision-making while maintaining interpretability and accountability. Our methodology retains the advantages of Bayesian optimization while incorporating expert knowledge and AI-driven guidance. The approach is demonstrated across various case studies, including bioprocess optimization and reactor geometry design, showing that even with an uninformed practitioner, the algorithm recovers the regret of standard Bayesian optimization. By including continuous expert-LLM interaction, the proposed method enables faster convergence, improved decision-making, and enhanced accountability for Bayesian optimization in engineering systems.
Establishing Multivariate Specification Regions for Incoming Raw Materials – a QbD approach
Establishing multivariate specification regions for selecting raw material lots entering a customer’s plant is crucial for ensuring smooth operations and consistently achieving final product quality targets. Moreover, these regions guide the selection of suppliers. By meeting these specifications, suppliers contribute to customer satisfaction, which can, in turn, enhance market share. Latent Variable Methods, such as Partial Least Squares regression (PLS) and the more recent Sequential Multi-block PLS (SMB-PLS), have proven to be effective data-driven approaches for defining multivariate specification regions. These methods model the relationships between raw material properties (Critical Material Attributes – CMA), process variables (Critical Process Parameters – CPP), and final product quality (Critical Quality Attributes – CQA), enabling the identification of a lower-dimensional latent variable subspace that captures quality-relevant variations introduced by raw materials and process conditions. This subspace is central to the methodology.
Within this latent variable space, several statistical limits are defined to ensure final product quality and guarantee data compliance with the latent variable model, collectively forming the multivariate specification region. After providing historical context on the early development of these methods, this presentation will cover several key topics. These include the data requirements for constructing latent variable models and how to organize them into distinct blocks. The techniques used to define the limits (in terms of shape and size) within the latent variable subspace—via direct mapping and latent variable model inversion—will be discussed, along with methods for addressing uncertainties. Particular attention will be given to how process variations generate different scenarios and approaches for establishing meaningful specification regions. Case studies using both simulated and industrial data will illustrate these methods. It will be demonstrated that the proposed framework aligns with the principles of Quality by Design (QbD), with notable similarities to the Design Space (DS) concept. The presentation will conclude by exploring potential future developments, including the establishment of multivariate process capability indices based on specification regions.