A flow vector assessment based on spiral thermal flow sensor and deep learning

Accurate flow measurement is essential for industrial, environmental, and biomedical systems that rely on precise control of fluid dynamics. Conventional flow-sensing devices tend to possess bulky form factors and intricate structural configurations, which can introduce disturbances to the flow field and limit their scalability toward miniaturized platforms. In this work, we present a monolithically integrated asymmetric thermal flow sensor (MATFS), realized through a selective NiO reduction method activated by laser irradiation. The process enables the seamless integration of a nickel heater and a NiO temperature sensor within a thin-film configuration, forming a compact and non-intrusive device. The spiral temperature sensor design around the heater captures flow-induced thermal asymmetry, allowing both flow speed and direction to be derived from resistance variations. This simplified hardware architecture, combined with a deep learning framework, enables accurate classification of flow vectors from a sensing element. The model effectively extracts spatial and temporal features from resistance data, yielding 100% accuracy for flow velocity estimation and 96% for flow direction classification. Furthermore, a wireless embedded module provides real-time monitoring and data transmission, enhancing the sensors applicability in dynamic environments. This approach demonstrates a scalable and versatile platform that integrates advanced laser processing, monolithic device fabrication, and machine learning-driven analysis, representing a significant step toward compact, intelligent, and high-precision flow sensing technologies.