Screening and optimization with one single OMARS design: design, analysis, and optimization
Conference Room 3
Irish College Leuven
Screening and optimization with one single OMARS design: design, analysis, and optimization
Part of the ENBIS-24 Leuven conference.
After the course, there will be a meet and greet with Peter Goos and Bradley Jones at 18h, and a drink offered by Effex at 19h.
Instructor
José Núñez Ares, EFFEX
Overview
The standard response surface methodology of conducting a screening experiment followed by an optimisation experiment has been challenged in recent years. In 2011, the emergence of a new type of design, the definitive screening design (DSD) (Jones & Nachstheim, 2011), posed a new challenge to this approach. In 2020, the first paper on OMARS designs (Núñez Ares & Goos, 2020) appeared, generalising DSDs and extending the use of a single design for screening and optimisation due to its orthogonality properties.
The selection of the best experimental design among different alternatives is crucial, on the one hand to minimise the experimental effort and, on the other hand, to maximise the information obtained once the experiment has been performed and the data collected. To achieve these goals, it is important to balance the size and different quality characteristics of the designs, such as projection estimation capacity, the power to detect different effects, or the number of replicate points.
The analysis of screening + optimisation experiments involves a large number of factors, making the analysis of experimental data difficult. Recently, a novel algorithm for all-subset model selection has emerged that can cope with problems with more than 100 potential effects and has been successfully applied to industrial problems (Vázquez, Schoen & Goos, 2020).
Optimisation of multi-response problems is often the end goal. The trade-off between the different responses and the high dimensionality of the input space (high number of factors) makes it challenging. The probability of success of being within specifications uses the predictive power of the underlying statistical models and quantifies the uncertainty and robustness of any given combination of factor values.
In this short course, we will guide the user through this process using our powerful yet intuitive design selection tool, new model selection algorithm and the optimization platform. This way, the best possible designs are coupled to the best possible models to analyze the precious data resulting from them.
Outline
The course is divided in two parts:
Part 1: how to choose an experimental design for screening + optimization
- Screening, optimization, or screening + optimization simultaneously?
- Quality attributes of a design and study of their trade-offs.
- Case studies.
Part 2: how to model experimental data involving a large number of factors and how to optimize multiple responses simultaneously
- All-subset model selection for problems with a high input dimensionality
- How to compare different models with each other
- Multi-response optimization problems and calculation of the probability of success
- Case studies
Pacticalities
- Each participant will receive a handout with the slides that will be used during the workshop
- Participants are invited to bring their own laptop. A trial account of the EFFEX software will be provided. Since it is cloud-based, no installation is required
- The instructor will guide the participant through all practical cases, in such a way that the participants will be able to replicate what the instructor does
Meet & greet with Peter Goos and Bradley Jones
After the course, there is a meet and greet with Peter Goos and Bradley Jones.
Drink in the city center
Participants are cordially invited for a drink at Leuven Centraal, Margarethaplein 3, 3000 Leuven. It is a 10 minute walk from the course venue.
Drinks start at 19h!
Short bio
José Núñez Ares was a postdoctoral researcher at KU Leuven's MeBioS research group (Mechatronics, Biostatistics and Sensors), where he studied ways to set up new, cost-efficient experimental designs.
He obtained a Master in Operations Research from Erasmus University in Rotterdam and a Bachelor in Civil Engineering from the University of Coruña (Spain). He obtained his PhD at KU Leuven under the supervision of Prof Peter Goos.
His research has been published in top journals and he is known in the academic community as the inventor, together with Prof Goos, of the OMARS design methodology. Besides his research, José is active in industrial consultancy and has successfully delivered integrated DoE solutions to companies in the pharmaceutical, chemical, energy and manufacturing sectors. He holds a US patent on an algorithm for experimental design selection.
José Núñez Ares is co-founder of EFFEX and assumes the role of Chief Scientific Officer.