Based on the assessment of each recycling technology, FIREFLY will propose various flowsheets to create and build a small-scale pilot. These technologies will be demonstrated in a predictive, RES-powered and flexible manufacturing process for new (electro)catalysts. Additionally, these new metal catalysts will be tested in innovative (electro)chemical processes of selected compounds.
Inlecom Innovation (INLECOM) leads the development οf the AI-based tool, which will predict the output of each individual electrochemical process and propose the optimal flowsheet for the recycling of spent catalysts, based on user requirements. After defining the requirements of the prediction tool and the system specifications, along with a preliminary assessment of potential machine learning (ML) algorithms, the team at INLECOM led a detailed system design. They also developed a prototype application featuring a user interface designed with a user-centric approach, in line with the specifications listed by the R&D partners.
The core of the development lies in the training and integration of ML algorithms. These algorithms are designed to predict the output and energy consumption for each individual technology within the electrochemical toolbox.
With the help of LPRC, researchers at INLECOM are currently equipping the toolbox with an interactive map enhancing the tool’s usability and accessibility. The map is founded on the data collected by LPRC within the GIS inventory.
The trained ML models for the electrochemical transformation in molten salts (ETMS) on TiO2 and the gas-diffusion electrocrystallisation (GDEx) for PGMs have already been integrated on the application, with the ETMS model already available in the application. The process of training optimal ML algorithms is computationally intensive and requires long processing times. Once the models are trained, additional post-processing corrections are applied to incorporate external knowledge into the model. This underlines the critical importance of human intervention in the optimisation phase of the dedicated ML algorithms created for these cutting–edge electrochemical processes.
The current status of the development is promising. The successful training and integration of ML models for ETMS on TiO2 and ongoing training for GDEx on PGMs mark significant milestones. The app development is progressing well, with the first version expected to be released to the consortium for feedback. Future work will involve further development and testing of the flowsheet optimisation algorithm, integration of additional ML models, and the completion of the interactive map functionality. The research team at INLECOM aims to achieve 80 % correctness in predicted results by the end of the project.