Nanoplasmonic hydrogen sensors for technologically relevant environments

Licentiatavhandling, 2025

The high flammability of hydrogen (H2) when mixed with air makes H2 sensors crucial in the development of hydrogen energy technologies across the Globe. Their development follows standards, in the form of performance metrics, with the most stringent and well-known ones established by the U.S. Department of Energy. Despite extensive research, key targets, such as the 1-second response time in 1000 ppm H2, remain challenging, often achieved only in ideal environmental conditions. Additionally, the influence of poisoning gas species (such as CO,CO2, NOx) in general and humidity in particular are rarely addressed. Among the plethora of different H2 sensing technologies nanoplasmonic optical sensors stand out as particularly promising. They rely on spectral shifts or intensity changes of the localized surface plasmon resonance of hydrogen-sorbing metal nanoparticles as the signal transduction scheme and have been boosted by advances in nanofabrication and the implementation of tailored nanostructured materials. Beyond materials engineering, advanced data analysis methods, such as the use of artificial intelligence (AI) that can greatly enhance data processing, are a to-date widely unexplored yet highly promising approach to improving the performance metrics of plasmonic hydrogen sensors. In this thesis, I apply both strategies and present three projects that aim to address both the response time and the humidity performance metric of plasmonic H2 sensors. In the first study (Paper I), we demonstrate a Pt-based catalytic-nanoplasmonic H2 sensor that can operate within 0-80% relative humidity (RH) and can detect concentrations as low as 600 ppm in air, at T≥ 50 ºC, while also exhibiting a decrease in the limit of detection with increasing humidity. In the second study (Paper II), we demonstrate the use of a tailored neural network ensemble model and showcase its ability to accelerate the response of a PdAu alloy nanoplasmonic H2 sensor by a factor of 40 by eliminating the H2 concentration/response time dependence. In the third study (Paper III) we employ a deep dense neural network to analyze data acquired from a PdAu alloy nanoplasmonic sensor operating under varying humidity (0-80% RH). With this AI-based treatment, we can eliminate the negative influence of H2O in the sensor’s response and achieve a limit of detection of 100 ppm at the highest measured 80 % RH.