数字孪生多模态视觉推理的神经-符号系统

Neural-symbolic system for multimodal visual reasoning towards digital twin

面对数字孪生在多模态视觉数据融合中的异质性和动态性挑战,提出一种结合深度学习与符号智能的方法。该方法通过深度神经网络对视觉数据进行实时解析,并借助符号系统存储的知识和事件响应规则,实现对复杂推理过程的自主管理。为提高系统对物理世界变化的适应性,提出一种融合多模态信息和外部知识的增强推理机制,该机制能有效地整合来自传感器的实时数据和历史知识库中的信息,以支持更加准确和合理的决策制定。以退役锂电池拆解过程为案例验证表明,该方法不仅能够在多模态数据环境中实现高准确率的识别和分析,还能够基于推理机制生成合理且逻辑一致的操作建议,有效提升了拆解效率和安全性。

Faced with the complexities of fusing heterogeneous multimodal visual data in digital twins,a novel neuro-symbolic approach for combining the analytical capabilities of deep learning with the structured reasoning of symbolic intelligence was proposed.This approach employed deep neural networks to analyze the visual data in real-time and supplemented autonomous management of complex reasoning processes by the knowledge and event-response rules stored in a symbolic system.To enhance the system's adaptability for the physical world changes,an augmented reasoning mechanism integrating multimodal information with external knowledge was proposed.This mechanism effectively consolidated real-time sensor data with information from historical knowledge bases to support more accurate and rational decision-making.The efficacy of the proposed method was demonstrated through a case study on the disassembly of retired lithium batteries,and its capability to achieve high accuracy in identifying and analyzing multimodal data was illustrated.Furthermore,the coherent and logical operational recommendations based on the reasoning capabilities were generated,which significantly improved disassembly efficiency and safety.