Education, Science, Technology, Innovation and Life
Open Access
Sign In
  1. Home
  2. Journals
  3. Automation and Machine Learning
  4. Vol 7, Issue 1, 2026

Lightweight Improvement of NeRF Algorithm for Industrial Digital Twin Scenarios in Automobile Factories

Download as PDF

DOI: 10.23977/autml.2026.070107 | Downloads: 1 | Views: 46

Author(s)

Xianrui Song 1, Xinyue Yan 1, Ting Zhang 2

Affiliation(s)

1 University of Sanya, Haikou, Hainan, 572022, China
2 Geely Automobile Research Institute, Ningbo, Zhejiang, 315300, China

Corresponding Author

Xianrui Song

ABSTRACT

Industrial digital twin technology provides core support for the intelligent operation and maintenance as well as flexible production of automobile factories. The Neural Radiance Fields (NeRF) algorithm, with its high-precision scene reconstruction capability, has become a key technology for scene modeling in automobile factory digital twins. However, the traditional NeRF algorithm suffers from high model complexity, slow inference speed, and high hardware deployment costs, making it difficult to adapt to scenarios with high real-time requirements in automobile factories, such as welding workshops and assembly lines. To address this pain point, this paper proposes a lightweight NeRF improvement algorithm (Factory-LiteNeRF) from three dimensions: scene partition modeling, network structure pruning, and feature encoding optimization, combined with the characteristics of typical digital twin scenarios in automobile factories. Experiments are conducted on the assembly line scene of an automobile factory to compare the model volume, inference speed, and reconstruction accuracy between the traditional NeRF and the improved algorithm. The results show that the improved algorithm compresses the model volume by 72.3% and increases the inference speed by 2.8 times, while ensuring that the reconstruction accuracy loss does not exceed 3%, which can meet the real-time modeling and operation and maintenance needs of automobile factory digital twins.

KEYWORDS

Automobile Factory; Digital Twin; NeRF Algorithm; Lightweight; Scene Reconstruction

CITE THIS PAPER

Xianrui Song, Xinyue Yan, Ting Zhang. Lightweight Improvement of NeRF Algorithm for Industrial Digital Twin Scenarios in Automobile Factories. Automation and Machine Learning (2026). Vol. 7, No. 1, 55-61. DOI: http://dx.doi.org/10.23977/autml.2026.070107.

REFERENCES

[1] Chen Gen. Digital Twin[M]. Beijing: Electronic Industry Press, 2020.
[2] MILDENHALL B, SRINIVASAN P P, TANCIK M, et al. NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis[J]. arXiv Preprint, arXiv:2003.08934, 2020. 
[3] Zhao Hui, Zheng Siyuan. Research on the Application of NeRF in the Communication Industry[J]. Information and Communications Technology and Policy, 2024, 50(12): 37-41.
[4] Dassault DELMIA Automotive Intelligent Manufacturing System: Flexible Production Line Planning and Factory Operation Efficiency Improvement Scheme Based on Digital Twin[R]. Chengdu Bestway Technology Co., Ltd., 2026.
[5] Project Case | A Digital Twin Platform for Automobile Production Lines in a Smart Factory[R]. Piaoshi Technology Co., Ltd., 2026.
[6] DING T, XIANG D, RIVAS P, et al. Neural Pruning for 3D Scene Reconstruction: Efficient NeRF Acceleration[J]. arXiv Preprint, arXiv:2504.00950v1, 2025.
[7] Liu H, Saksham D, Shen M, et al. Review of Digital Twin in the Automotive Industry on Products, Processes and Systems[J]. International Journal of Automotive Manufacturing and Materials, 2025, 1(1): 1-25..
[8] Lingtu Interaction. 5G Automobile Factory Digital Twin Collaboration Scheme[R]. Lingtu Interaction (Wuhan) Technology Co., Ltd., 2026.
[9] BMW Group. BMW iFACTORY Digital Twin Technology Application Report[R]. BMW Group, 2025.
[10] Ford Motor Company. Practice of Digital Twin Technology for Equipment Operation and Maintenance in Automobile Factories[R]. Ford Motor Company, 2025.
[11] Soori M, Arezoo B. Digital twin for smart manufacturing in automotive industry[J]. Journal of Manufacturing Processes, 2023, 96: 413-430. https://doi.org/10.1016/j.jmapro.2023.05.012.
[12] BARRON J T, MILDENHALL B, TANCIK M, et al. Mip-NeRF: a multiscale representation for anti-aliasing neural radiance fields[J]. arXiv Preprint, arXiv:2103.13415, 2021.
[13] MÜLLER T, CHAURASIA G, FRITZ C, et al. Instant neural graphics primitives with a multiresolution hash encoding[J]. arXiv Preprint, arXiv:2201.05989, 2022.
[14] Digital Twin Visualization System for Automobile Manufacturing Workshops Based on HT Technology[R]. Tutu Software Co., Ltd., 2026.
[15] Google Brain. MobileNetV3: Efficient Convolutional Neural Networks for Mobile Vision Applications[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(10): 2440-2451.

Subscription
E-Mail Alert
Most popular papers
Call for Special Issue
Publication Ethics & OA Statement
Plagiarism Policy
Contact Us
Downloads: 4905
Visits: 242216

Sponsors, Associates, and Links

All published work is licensed under a Creative Commons Attribution 4.0 International License.

Copyright © 2016 - 2031 Clausius Scientific Press Inc. All Rights Reserved.