基于神经渲染的数字孪生资产快速场景几何建模与检索方法

Neural rendering-based fast scene geometry modeling and retrieval method for digital twin assets

虚实融合是数字孪生技术的典型特征,数字孪生的虚拟场景大多通过几何建模技术来实现。为解决场景几何建模的自动化程度较低。过于依赖人工,导致成本高昂且效率低下的问题,提出一种数字孪生几何场景构建方法。引入神经渲染技术采集物理实体的点云数据,然后设计一种基于深度学习的点云模型与3DCAD模型语义映射的方法,用于基于数字孪生几何模型资产库的3DCAD模型检索。最后建立训练数据集并构建数字孪生几何模型资产库进行实验验证。通过对比实验和退役电池拆解案例验证:与其他的数字孪生场景几何建模方法相比,该方法具有更低的成本和更高的效率。

Virtual-real fusion represents a quintessential facet of digital twin technology.In digital twins,virtual scenes are predominantly realized through geometric modeling techniques.To address the challenge of limited automation and excessive dependence on manual intervention in scene geometric modeling,resulting in high costs and inefficiencies,a digital twin geometric scene modeling approach was proposed.In this approach,the neural rendering technology was introduced to gather point cloud data from physical entities.Subsequently,a deep learning-based method was devised for the semantic mapping of point cloud models to 3D CAD models,which was applied to retrieving 3D CAD models from the digital twin geometric model asset library.A training dataset was curated,and a digital twin geometric model asset library was constructed for experimental validation.The efficacy of the proposed approach was corroborated through comparative experiments and case studies involving the disassembly of decommissioned batteries.The results affirmed that the proposed method exhibited significantly lower costs and markedly heightened efficiency than alternative digital twin scene geometric modeling methodologies.