An Intelligent Reinforcement Actuating Sensing Learning System Based on the Neuromorphic-Induced Electrodynamics Model

In an effort to transcend the limitations of conventional eddy current sensing systems that have thus far been confined to demonstrating its capabilities within a constant sensing strategy to passively obtain digital datasets, this article proposes a prototype of an intelligent eddy current sensing system based on the neuromorphic-induced electrodynamics model rooted in the physical domain. Within this system, a physical electromagnetic (PEM) neural network has developed by employing transmitters and receivers as its physical neural entities, enabling the sensing system to directly and intelligently process the physical properties in the real physical world. Furthermore, a physics-informed reinforcement learning (RL) framework has been harnessed to dynamically manipulate the weights of the PEM network, ensuring the adaptive maintenance of optimal sensing conditions in evolving environments. Through this approach, the proposed sensing system implemented a self-decision-making mechanism and generated a self-curated optimal physical dataset and corresponding labels. Through a specific application in nondestructive material testing, the adaptability of the proposed system in achieving optimal sensing conditions in varying environments is validated. This article lays the potential paradigm shift that allows artificial intelligence (AI) to extend its scope from the digital domain to the real physical domain and may open up more applications, such as smart sensors, data generation, and online inference tasks.