ALCHIMIA interview with the coordinator
As far as understanding the mission that the ALCHIMIA Project is undertaking, there is no better place to start than an interview with the coordinator, Jesus Benedicto Cirujeda. Here, he talks about what ALCHIMIA is trying to achieve, what distinguishes it from other projects, as well as opening up the bonnet and exploring the inner workings of the engine that is driving a project at the forefront of the European green energy industry:
In simple terms, what is the ALCHIMIA project trying to achieve?
In simple terms, ALCHIMIA aims to optimize the process of producing steel from scratch so that it becomes greener and supports the green deal. With the use of artificial intelligence techniques, the idea is, basically, to improve these industrial processes by reducing energy consumption, emissions and waste production.
How is the ALCHIMIA project different from other research projects in the same field?
What differentiates the ALCHIMIA project from other research projects in the same field could be summarized in the initial value proposition we have just identified as part of our exploitation strategy. Economical savings, reduced energy consumption, reduced carbon footprint and waste are the basis of the benefits ALCHIMIA expects to offer their users. ALCHIMIA project is the ideal meeting point for industries to collaborate; the generated synergies would lead to technological improvements and advances in the research fields. Industry associations will be able to improve the economic and environmental performance of associated companies by offering support for the deployment of the ALCHIMIA platform. Even, the federated learning approach followed in the project could push within industry clusters the improvement of the platform itself and this may represent a new asset for industry associations. Apart from that, ALCHIMIA platform opens a window to the creation of new standards and test-environment for policies. Finally, but not less important, ALCHIMIA project has a noticeable environmental impact and plant productivity will be improved with a positive economic impact on society. Our approach to achieving this goal prioritizes the human element by ensuring that workers understand how new AI systems will complement and enhance their capabilities and roles while providing trust and safety.
What are the main challenges that the ALCHIMIA project aims to address, and how does ALCHIMIA plan to tackle them?
In general, ALCHIMIA project will address the challenges of the steel sector, creating an innovative system that automates and optimises the production process dynamically with a holistic approach that includes scrap recycling and steelmaking. ALCHIMIA will find an optimal mix to reduce energy consumption, emissions and waste generation of steelmaking while guaranteeing to obtain high-quality products.
If we go a little deeper, within the ALCHIMIA project, a Federated Learning framework will be specifically developed to address the requirements of the metallurgy sector, where ML models must be developed without sharing data even between factories that belong to the same group; the challenge here is whether this framework could be used in various industries with similar model quality and deployment speed. ALCHIMIA will provide multiple aggregation algorithms which will be integrated into the solution to adjust the performance to different use-cases; besides that, local adaption and Transfer Learning techniques will be applied to fight against accuracy problems of federated models.
During the deployment of machine learning models in production environments, significant drops in production may occur several times when input data is slightly different with respect to the training datasets; the challenge here is to develop and extend methods and tools to monitor in situ the neural network error analysis and to bench the robustness in order to detect the problem. ALCHIMIA will implement and demonstrate a continual learning paradigm so that models’ performance can be monitored once they are deployed within systems on the shop floor, triggering retraining processes to overcome domain shift issues and improving the robustness of neural networks.
Although artificial intelligence models can automate several tasks, some concerns have been raised during the last years due to the uncertainty that their usage can cause in critical domains like manufacturing. To overcome this challenge, ALCHIMIA will use the last guidelines, recommendations and emerging regulations from relevant European initiatives to guarantee trustworthy artificial challenge, considering that in some cases the usage of artificial intelligence can be classified as high-risk. ALCHIMIA will assess all the technologies and use-cases considering the ALTAI questionnaire and the EC proposed regulation so that the project becomes one of the first pilots for their adoption.
Today the material feedstock characterisation is mainly based on simple multiple linear regression calculations; however, information on the standard deviations is very difficult to assess by these standard methods. To solve this, in ALCHIMIA the statistical analysis methods will be connected to process databases and coupled with tools for the automatic updating of material property parameters. Also, innovative artificial intelligence and big data analysis methods will be applied for more accurate and up-to-date parameter estimation. These enhancements will make possible to track also short- and medium-term variations in feedstock material quality, and to determine the properties of new or seldomly used charge materials. The continuously and automatically updated information on material properties will be fed to charge mix optimisation tools and model-based process monitoring and control systems, to ensure that at any time optimal solutions can be provided to the operators.
One of the main objectives of the ALCHIMIA project is to improve the determination of the material characteristics regarding composition and energy demand. More precise capture of these characteristics would lead to significantly higher model accuracy. Furthermore, the existing first principle process models will be enhanced by implementing artificial intelligence solutions for online parameter estimation and optimisation. This hybrid approach will ensure the long-term stability and robustness of the process models while improving the accuracy and conserving the interpretability of the results.
These days, the vast majority of automatic control loops in the process industries still rely on various forms of PID controllers. In recent years, Model Predictive Control systems, the called MPC, were emerging to provide a more intelligent control strategy. ALCHIMIA pursues a green MPC concept for minimal environmental impact by including the life cycle assessment in addition to the steel quality targets in the optimisation problem of the MPC. In a first step, the results will be displayed in the form of optimised control suggestions for an operator in order to build trust in the artificial intelligence framework and hybrid models developed in ALCHIMIA. Based on the experiences of the users it can be considered to utilise the closed-loop control capabilities of the MPC to further optimise the process.
Artificial intelligence applications and machine learning can result in opaque processes and outputs from the point of view of workers; ALCHIMIA will produce a set of recommendations which will provide insights and foresight recommendations for humancentric technology development and insertion. Such recommendations will be complemented by a skills gaps assessment and training recommendations based on good practices and existing courses. Overall, the project intends to produce an Industry 5.0-based toolkit for addressing social aspects in dealing with technological innovation.
Finally, we cannot forget that with the increasing socio-economic and environmental problems, making human activities sustainable has become a vital challenge for the future of our societies. ALCHIMIA will produce an sustainable development goals assessment, identifying both the potential negative and positive impacts of a project on the sustainable development goals and translating the results into a practical action plan to better align their activities with the sustainable development goals and the EU Green Deal. Beyond its formal results, ALCHIMIA will also contribute to raising awareness of the sustainable development goals requirements through participation in relevant events, workshops and webinars.
How does the ALCHIMIA project integrate different disciplines and expertise to achieve its goals?
ALCHIMIA consortium is composed of 10 entities with proven commercial activity in ICT, AI and manufacturing. Among these entities, we can find industrial big companies and SMEs, research technology organizations and academic research departments from universities. This heterogeneous group provides high competencies in fields such as big data, federated learning, transfer learning, modelling, optimisation, life cycle assessment, trustworthy and explainable artificial intelligence, industrial internet of things or secure shell; among the ALCHIMIA members, there are manufacturing enterprises which offer the project the possibility to get access to large-scale and real-world industrial production lines and industrial facilities. And besides that, some consortium members have deep expertise in environmental impact assessments. If this were not enough, ALCHIMIA benefits from international cooperation with one Swiss self-funded associate partner. As can be seen, ALCHIMIA operational capacity covers the full spectrum of technologies and know-how required to reach the project goals.
What are some of the expected outcomes of the ALCHIMIA project?
Several outcomes are expected from this ambitious project. The first one is to implement a decentralized Artificial Intelligence and data solution to help the European big metallurgy industries become greener; this solution, and this is another expected outcome, will find the optimal mix needed in steel-making processes based on recycled scrap metal in an automatic and dynamic way; this would reduce the environmental impact of these industrial activities. Apart from that, ALCHIMIA will guarantee the highest levels of trust, safety and seamless collaboration between workers and AI-powered industrial solutions, something critical when we are talking about collaboration between different entities. Other expected outcomes are related to the establishment of synergies with other artificial intelligence on-demand EU platforms, such as GAIA-X, and the elaboration of a proper communication and exploitation strategy which allows ALCHIMA results to be adopted in several industrial sectors.
How does the ALCHIMIA project contribute to broader efforts to address environmental and societal challenges?
The main aim of the project is a statement of intentions itself: The development of an innovative artificial intelligence, data and robotics solutions for resource optimisation and minimisation of waste. Apart from that, ALCHIMIA innovative solution plans to reduce energy consumption, at least, by ~5% and scrap metal waste by ~10% for production of 1 Tn of steel; talking about gas emissions, the increased process and resource energy would be able to reduce them in another 5%. But our green digital solution developed in ALCHIMIA will not only allow us to reduce the impact of industrial companies on the environment; it will be respectful of the workers involved in those companies. They will be involved in a skill development program which motivates them to be an important active part of the future digital factories. Thanks to those skill programs, workers will increase their knowledge of the factors affecting the environmental performance of the processes they are part of, and they will feel like they are really part of the challenge of mitigating climate change through their work.