Veröffentlichungen und Präsentationen

This work introduces an AI-based predictive tool designed to optimize decision-making for the electrochemical recycling of spent catalysts. Developed within the FIREFLY project, the tool optimizes complete, multi-step metal recovery flowsheets. These flowsheets comprise advanced electro-driven technologies like Gas-Diffusion Electro-crystallization (GDEx)[1], Electroleaching (ELX)[2] and Electrochemical Transformation in Molten Salts (ETMS)[3]. Users can define the optimization criterion of choice (e.g. maximization of recovery efficiency or minimization of the total energy consumption) as well as additional constraints for the system to uphold. The foundation of this AI-based predictive tool is surrogate Machine Learning (ML) models for each cutting-edge electrochemical technology, trained on high-fidelity datasets. Since data quality is of utmost importance for training dependable ML models, massive datasets were generated using COMSOL Multiphysics® to map operational variables (like cell voltage or current density) to target outputs like metal recovery efficiency. For example, the GDEx process was modeled[4] with a dataset comprising 2,375 simulated experiments (1,233,220 individual data points). Multi-Layer Perceptron (MLP) artificial neural networks were trained as surrogate models for processes like GDEx and ETMS for which large datasets were available. To ensure robust generalization and prevent data leakage, a specialized data-splitting strategy was employed, dividing data by entire experimental runs rather than randomized data points. Various ML and deep learning models were evaluated, with the MLPs demonstrating the best performance. For example, the GDEx surrogate model achieved an exceptional R² score > 0.99 and a Mean Absolute Percentage Error (MAPE) < 0.5% on the test set. Processes with smaller simulated datasets (like the ELX process, where ML algorithms would likely overfit and fail to generalize) were modeled utilizing physics-informed analytical functions. The culmination of this single-technology modeling is the integration of these surrogate models into a comprehensive flowsheet optimization engine. Hosted on a ReactJS web application and powered by a Neo4j Knowledge Graph database, the platform connects the individual ML and analytical models into sequential processing chains. Using gradient descent algorithms, the tool evaluates the parameter space across the entire flowsheet to identify the configuration that satisfies the user-defined criteria. For instance, it can determine operational parameters that minimize total energy consumption while maintaining a recovery efficiency above 90%. This AI-based predictive decision support tool allows complex flowsheets to be evaluated and optimized within minutes to hours before any actual pilot testing, guiding operations toward the optimal configurations. Ultimately, the platform bridges the gap between computationally heavy multiphysics simulations and near real-time industrial application, providing scalable decision support for the circular economy infrastructure.

The FIREFLY project has received funding from the European Union’s Horizon Europe research and innovation programme under Grant Agreement No 101091715. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Health and Digital Executive Agency (HADEA). Neither the European Union nor the granting authority can be held responsible for them.