Physics-informed neuromorphic learning: Enabling scalable industrial digital twins

Digital twins (DTs) have shown promise in industrial automation by facilitating seamless fusion between virtual and physical spaces, thereby enhancing operational efficiency. However, existing DTs are predominantly customized implementations, requiring advanced expertise and significant resources, which hinders their broader application. Here, we propose a physics-informed neuromorphic learning framework, a computationally efficient approach based on the brain-inspired structural representation. This physics-guided representation, inferred and integrated through prior cross-modal structural knowledge, provides a scalable foundation for DT construction with brain-like generalization capabilities and ultra-low computational complexity. We demonstrate its effectiveness across multiple industrial scenarios, attaining accuracy comparable to current DT modeling techniques while reducing computational latency to less than 1/30. This framework shifts the paradigm from complex, customized DTs to a simplified, unified solution, potentially saving considerable human and computational resources. By facilitating the practical integration of DTs into industrial workflows, our method marks a substantial advancement in accelerating industrial transformation.