​Hosted by the Design of Experiments Group, KU Leuven

​We invite you to a connected series of talks organized by the DoE Group of Prof. Dr. Peter Goos 🔗. Rather than three independent seminars, this afternoon is designed to highlight different but complementary perspectives on Experimental Design. The session is organized and presented by PhD researchers Jade Lejeune Herman 🔗, Robin van der Haar 🔗, and Ying Chen 🔗.

​📍 Format: Hybrid (In-person at KU Leuven & Online)

​ℹ️ Note on Attendance: Participants are welcome to attend the full seminar or to join individual talks, depending on their availability. The afternoon is modular, and you are welcome to enter or leave during the breaks between talks.

​🗓️ Programme Overview

​14:30 – 15:10 | Scaling of Design of Experiments: A Genetic Algorithm for Large Orthogonal Designs

​Speaker: Jade Lejeune Herman

​High-throughput experimentation in modern product development presents new opportunities and challenges for experimental design. While traditional Design of Experiments (DOE) methods extract maximum insight from limited tests, they do not scale well to robotic platforms capable of massive parallel experimentation. This paper presents an AI-driven Genetic Algorithm (GA) framework for constructing large-scale Orthogonal Minimally Aliased Response Surface (OMARS) designs with uniform precision. These three-level designs ensure orthogonality among main effects, two-factor interactions and quadratic terms, enabling accurate modeling of nonlinear systems. Using constraint-aware operators and fitness-guided evolution, the GA efficiently searches vast design spaces to identify high-quality orthogonal or near-orthogonal solutions. The framework successfully generates OMARS designs with up to 96 runs and 12 factors, substantially exceeding previous constructions and enabling scalable DOE for automated, high-throughput research and industrial applications.

​

​15:10 – 15:50 | Design of Experiments under Time and Cost Constraints: A Real-World Case Study from the Potato Fries Industry

​Speaker: Robin van der Haar

​Textbook experimental designs often fall apart the moment they hit the factory floor, where logistical realities like cooling times, cleaning protocols, and strict shift schedules override statistical theory. Using a pilot-scale production line for Belgian fries as a case study, we illustrate how factors such as temperature and residence time create complex cost constraints, where heating times depend on vessel volume and the order of runs is dictated by thermal inertia. These complexities render traditional DoE methods ineffective.

​We propose a generalizable, modular framework that allows practitioners to model these "messy" industrial constraints and integrate them directly into the design optimization process. By moving beyond rigid structures and fixed run sizes, we demonstrate how to generate designs that are both statistically efficient and operationally feasible. 

​Whether producing potato fries or managing pharmaceutical reactors, this methodology ensures that experiments fit the available budget without sacrificing scientific rigor.

​

​15:50 – 16:30 | Optimal Design of Experiments for Powerful Equivalence Testing

​Speaker: Ying Chen

​Robustness studies in biopharmaceutical and biological manufacturing are routinely conducted to demonstrate that, under normal operational variability, process responses remain equivalent to those at nominal operating conditions. While equivalence is the primary inferential objective, commonly used experimental designs are typically optimized for estimation or prediction and can therefore provide inadequate power for equivalence testing. We propose a new semi-Bayesian optimal design criterion, termed powerful equivalence (PE) optimality, for studies intended for equivalence assessment via the Two One-Sided Tests procedure. PE-optimality maximizes the probability of correctly concluding equivalence across the experimental region. This criterion provides a statistical framework for both design selection and run size determination. We show that PE-optimal designs consistently attain higher regional equivalence power than conventional optimal designs for a fixed run size. We illustrate the methodology with a robustness design for a vaccine adjuvant manufacturing process.

​🤝 Why a joint seminar?

​While each talk addresses a distinct topic, they are intentionally grouped to encourage cross-fertilization between different DoE approaches. We hope this format will be of interest both to those working deeply within one area and to those curious about adjacent perspectives.