Movers and Shakers with Claes Holmstrom, CEO at Ferritico

While technology is making inroads into every industry, Material R&D may have been late to the party. While there may still be hurdles, the good news is opportunities are beginning to be realised.

We spoke to Claes Holmstrom, CEO at Ferritico to pick on his mind and understand his perspectives on the industry and the current ecosystem

reogma: How is technology disrupting the material R&D space and please give a few examples of such tech (AI, ML, Digital Twin, BD etc) and their impact?

The potential in AI for materials R&D is that novel alloy recipes can be developed inside a digital environment as opposed to a physical. The reduced need of the conventional time- and resource consuming physical trial-and-error approach ramps up R&D-tempo significantly and can shorten lead-times for novel steel development from years to months.

It might sound visionary and even illusionary for someone working in a steel mill today, but we believe materials customization could be a reality when AI and simulation driven development is fully implemented. Imagine the disruptive force and the societal impact of product designers going from selecting good enough materials being available off-the-shelf to being provided a customized and optimized material for specific applications.

In contrast to conventional physical modelling based materials simulation software, AI includes generative features and provides the capability to generate optimization candidates, i.e. steel recipe candidates are generated based on an inputted material properties prioritization order.

The generative approach constitutes the major technology leap for materials R&D since it does not only make the process engineer faster when mimicking the current experiment based and iterative approach, but makes him smarter since starting to iterate closer to the goal.

reogma: What are the regional under currents in your observation that are driving tech inroads? Which regions/countries are ahead in terms of innovation, adoption, use cases etc?

We see regional differences when it comes to IP protection in terms of willingness to share and collaborate on data. The data policies for steel and manufacturing companies vary but in general there is still a protectionist sentiment and poor understanding about what is actually revealed if sharing models and not underlying raw data, using a federated learning approach.

For the benefit of industrial AI roll-out, we see correlation between organizational adaptation of AI and reluctance in sharing data, i.e. when not understanding the underlying technology you rather say no than taking risk on enabling competitors to derive sensitive IPs through shared AI models. Hence, we see that data collaboration increases as the general industrial maturity of AI increases. 

reogma: Are there any open source or interoperable platforms (that exist) to drive application scalability? Is there an industry need for such collaboration?

There are many good initiatives although still at an early stage of development and roll-out. To mention one, I like the ambitions of the Zero Defects Manufacturing Platform (ZDMP) project supporting deployment and integration of simulation features for individual material properties and subprocesses provided by multiple different vendors.

The end AI goal and the major materials simulation business case is based on materials value chain integration, i.e. that data is shared centrally between multiple parties along the integrated material life cycle and to enable modelling and optimization over the entire chain, in contrast to only modelling and optimizing the individual steps as today. There is a need for software platforms that support this but technical as well as IP related challenges make the progress slow. 

reogma: What is your role in this market? How are you participating? What is your differentiator?

There are many actors out there that provide state-of-the-art machine learning and AI services. Our strength is our domain knowledge and having our background in materials and specifically steel research, focusing on materials modelling at different length scales.

We believe the key success factor in modelling materials is the fact that AI is not the one and only solution, but to consider it as a key part of an Integrated computational materials engineering (ICME) toolbox where we cherry pick the best performing solution for each individual simulation task, independently of whether it is physical modelling, first principles calculation or machine learning. Data is a scarcity in our industry and knowing the pros and cons of conventional simulation technologies and combining it smartly with AI to mitigate the data challenge is our secret sauce

In general, steel and manufacturing companies are currently implementing extended process data collection and warehousing capabilities to enable process optimization through data modelling using digital twins.

These data infrastructures provide the companies sufficient data volumes within a narrow data space, i.e. the data intervals are slim since it derives from natural deviations in trimmed processes. Hence, it provides them the possibility to model, view correlations in process variables and optimize within slim data ranges, while being without simulation support when considering significant process engineering, e.g. upscaling a novel steel grade or evaluating a significantly different heat treatment process.

Hence, there is a request for our generic machine learning models, based on scattered alloy compositions and wide process data ranges, to support the local and specialized digital twin models when considering major process engineering. This field is immature and there are still technical challenges related to the integration of digital twin and global models. We do however believe this will be prioritized as the industry wants to leverage the full value of their digital twin implementations in the years to come.  

About Ferritico

Ferritico is a machine learning-based steel simulation SaaS that makes the steel development and manufacturing processes more efficient.

Trial-and-error steel development can be time and resource consuming. Therefore, we have created the Ferritico steel simulation SaaS, where metallurgical know-how, large databases and machine learning are combined to help you predict various steel characteristics.

The long experience of the Ferritico team from metallurgical research, materials modelling, machine learning, and digitalization, has been instrumental for the development of the Ferritico SaaS. The service is currently being developed with some of the leading steel producers in Europe with the purpose of helping them make their steel development and manufacturing processes more efficient. The mission of the Ferritico team is to help digitalize the steel industry. 

Ferritico offers you a set of powerful statistical models to help you predict the characteristics of steels that you are developing. Currently, you can use the service to predict important temperatures such as Ms, Bs, Ac1, Ac3 as well as the martensite fraction (retained austenite) and the martensite hardness. More models are under development and will soon be available.

Leave a Reply

Next Post

Fuel Cell Opportunities in Japan’s Automobile Market

Mon Jan 17 , 2022
Japan is backing up hydrogen technology attuned to its ambition to minimize harmful emissions and help ensure energy security. It is expected that Japan’s automotive fuel cell market will accelerate at CAGR of 30% and reach $1510 by the end of 2030. Cost, infrastructure and low level of end user […]