Artificial neural network reinforced topological optimization for bionics-based high-performance hydrogen sensor design under multi-physical field coupling

Hydrogen energy is considered a promising secondary energy source due to its green and efficient nature, earning it the title of the ultimate energy source of the 21st century. For security, resistive hydrogen sensors are a favored and well-established class of sensors developed to detect leaks during the production, transportation, and storage of hydrogen. However, the lack of focus on designing and optimizing hydrogen sensing arrays has limited the sensitivity of these devices. In response to this limitation, aided by topology optimization reinforced by artificial neural networks, a bionic tridimensional columnar hydrogen sensor (TCHS) has been developed based on multi-physics field coupling. Inspired by a paper-fan-like structure, TCHS has demonstrated a 70 % increase in strength compared to conventional rectangular arrays. Extensive simulations using back-propagation neural networks have validated the TCHS. Unique sensing cell structure allows the sensor to monitor hydrogen leaks from 0.1 % to 4 % at room temperature (25 °C), with a maximum response of 23 %, showcasing its superior electrical performance. This research marks a significant advancement in the performance of hydrogen sensors widely used in early warning devices, gas detection, and biosensors, providing new theoretical guidelines for their enhancement.