A Deep-Learned Monolithic Nanoparticle Asymmetric Thermal Flow Sensor for Flow Vector Estimation

Abstract

Flow sensing is essential in various fields, including industrial, environmental, and biomedical applications, where accurate measurement of fluid dynamics is crucial. Traditional flow sensors are often bulky and complex, which can distort the flow and complicate installation when placed directly in the flow path. To address these issues, we developed a deep-learned monolithic asymmetric thermal flow sensor. The sensor is fabricated via laser-induced selective sintering and reduction of nickel oxide nanoparticles, seamlessly integrating a microheater and temperature sensors into a thin-film device. This thin-film design minimizes flow disturbance and improves measurement accuracy. Unlike conventional calorimetric flow sensors that require complex multiarray electrode configurations, our system features a temperature sensor designed in an asymmetric spiral shape around the heater. This optimized hardware configuration not only simplifies the structural design but also supports deep learning algorithms for accurate flow estimation. By integrating this asymmetric design with reinforcement learning algorithms, the sensor efficiently bridges hardware and software, enabling precise flow vector estimation based on changes in sensor resistance. Furthermore, equipped with an embedded wireless communication system for real-time data monitoring, the sensor ensures reliable flow assessment, making it a versatile solution for diverse flow estimation applications.