Accelerated quantification of photolytic hydrogen using multi-fidelity Bayesian optimization and automation

Descrizione del progetto

Accelerare il processo di fotolisi per ottimizzare la resa dell’idrogeno

L’idrogeno verde, un promettente sostituto ecocompatibile dei combustibili fossili in numerosi settori industriali, può essere prodotto attraverso la scissione fotocatalitica dell’acqua utilizzando come (foto)catalizzatori varie strutture organiche inorganiche, nobili o nano-covalenti; ciononostante, al fine di ottimizzare la resa di idrogeno è necessario prendere in considerazione numerosi fattori, come la selezione dei co-catalizzatori, i rapporti tra catalizzatori e co-catalizzatori e i livelli di pH e viscosità adeguati. Testare manualmente molteplici combinazioni di parametri richiede molto tempo, mentre i laboratori di guida autonoma possono sfruttare la robotica avanzata, la potenza di calcolo e l’intelligenza artificiale allo scopo di ottenere risultati in maniera molto più rapida. Sostenuto dal programma di azioni Marie Skłodowska-Curie, il progetto SDL-MFHYD si prefigge di accelerare efficacemente il processo di fotolisi oltre le capacità attuali basandosi sull’impiego di un algoritmo di ottimizzazione bayesiana multi-fedeltà che consente di ridurre la frequenza delle fasi critiche dispendiose in termini di tempo nella fotocatalisi.

Obiettivo

The fossil fuel sector is projected to emit 200 million tons of CO2 equivalent by 2050. Hydrogen is emerging as a crucial energy carrier, essential for achieving net-zero emissions (NZE) by 2050. The European Commission is actively funding initiatives for decarbonization and green hydrogen production. Green hydrogen can primarily be produced through photocatalytic water splitting, involving either proton reduction or overall water oxidation. While several photocatalysts, predominantly inorganic or noble materials have been reported, recent advances in environmentally friendly nano-covalent organic frameworks (Nano-COFs) catalysts offer tunability and significant synthetic diversity. However, photocatalysts alone are insufficient for substantial hydrogen production. Multiple components must be integrated, such as co-catalyst selection, catalyst-to-co-catalyst ratios, and physicochemical parameters like pH and viscosity, to optimize hydrogen yield. The complexity of optimizing these parameters is challenging for manual testing, especially as the search space expands exponentially. Self-driving laboratories (SDLs) are poised to revolutionize this field by leveraging advancements in robotics, computational power, and artificial intelligence (AI). SDLs can achieve scientific objectives hundreds of times faster than traditional automation, integrating hardware for experiment execution and software for data analysis and subsequent experiment design. Despite these advancements, the time-intensive steps of photolysis and gas analysis remain bottlenecks. This proposal addresses the challenge of accelerating the photolysis process beyond current SDL capabilities. By employing a multi-fidelity Bayesian optimization algorithm, I aim to reduce the frequency of crucial yet time-intensive steps in photocatalysis. This novel approach, untested in real photolysis experiments, has the potential to extend broadly to other areas of electrochemistry, including CO2/N2 electrolysis.