Berlin 2024 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
O: Fachverband Oberflächenphysik
O 46: Plasmonics and Nanooptics IV: Fabrication and Applications
O 46.5: Talk
Wednesday, March 20, 2024, 11:30–11:45, MA 042
Accelerating Plasmonic Hydrogen Sensors by Transformer-Based Deep Learning — •Viktor Martvall1, Henrik Klein Moberg1, Athanasios Theodoridis1, David Tomeček1, Pernilla Ekborg Tanner1, Sara Nilsson1, Giovanni Volpe2, Paul Erhart1, and Christoph Langhammer1 — 1Department of Physics, Chalmers University of Technology, Gothenburg, Sweden — 2Department of Physics, University of Gothenburg, Gothenburg, Sweden
Fast and accurate H2 sensors are needed for H2 technologies to address safety concerns associated with the high flammability of H2-air mixtures. Plasmonic optical hydrogen sensors, monitoring H2 through changes in the localized surface plasmon resonance peak of metallic nanoparticles absorbing hydrogen, shows promise. In idealized H2-vacuum conditions, they have met the US department of energy’s target of a response time < 1 s for concentrations < 0.1 vol.% H2. However, further advances are required to meet this target in a realistic environment, where the presence of other molecules slows down the sensor response. Here, we accelerate sensor response by developing a deep learning (DL) model to predict the H2 % from the time-dependent extinction spectrum. We apply the DL model to a Pd70Au30 alloy plasmonic sensor in Ar carrier gas. Compared to the conventional analysis, collapsing each spectrum to a single spectral descriptor related to the H2 % via a power law, our model demonstrates up to a 40 times faster sensor response time during rapid H2 pulses. Also, we illustrate that it can faster discern and quantify gradual changes in H2 %.
Keywords: Hydrogen sensing; Localized surface plasmon resonance; Deep learning