“Essentially, all models are wrong, but some are useful.” Quoting statistician George Box, Ph.D. Michele Mattei of AkzoNobel challenged the academic researchers in his presentation at the Discussion meeting.
“Scientists generally strive to create models that have extremely high accuracy. But from an industry point of view, usability is more important than accuracy,” he elaborates.
Smilingly, he notes that he recently was a CERE scientist himself. Two years ago he finished his PhD-project which was conducted at CERE and CAPEC, another center at DTU. Returning to CERE for the Discussion meeting he gives the first AkzoNobel presentation at the event.
“We are happy to be part of the CERE industry Consortium. Not just to follow the research at CERE, but also to contribute actively by providing real industry challenges for them to address,” says Michele Mattei.
As a leading producer of paint and coatings, and a major producer of a range of specialty chemicals, AkzoNobel supplies essential ingredients, essential protection and essential colour to industries and consumers worldwide.. The company has more than 45,000 employees in more than 80 countries. There are 4,000 scientists and technologists; Michele Mattei is in specialty chemicals research.
“This is the field closest to the activities of CERE. Since AkzoNobel joined the Consortium in 2002, the tendency has been for CERE to lean towards oil and gas, but the activities are still highly relevant for the specialty chemicals industry. A number of the issues we have in process design and control are very similar to the oil and gas industry, and the same solutions can often be applied.”
As an example, Michele Mattei points to electrolyte thermodynamics. CERE has developed new models and software in this field which is highly relevant to AkzoNobel, as the company is the world leader in refined salt.
“One thing I like a lot in CERE’s efforts is the balance between modeling and experiments. You need to constantly link both types of developments. When experiments and model predictions differ we tend to always blame the model. This is really not fair. Both experiments and models contribute to uncertainty. You need to always keep the full picture in mind,” Michele Mattei states, returning to the key point from his presentation: “Rather that striving for a model that is fully predictive, thus not requiring any experiment, from an industrial perspective it is more productive to use semi-predictive approaches, where few dedicated experiments can be used to extrapolate results to a wider range.”
“Combining experiments and modeling results with uncertainty indications, however, is in general gaining interest in industry, and my recommendation to CERE would be to also pay attention to this development.”