Image-to-CAD Reconstruction for the Digital Delivery of Power Equipment: Validation Across Four Insulator Types

Digital delivery of power equipment requires integrating design information and the physical geometry of individual manufactured units into a single computer-aided design (CAD) model. To address the time-consuming nature of unit-by-unit measurement and manual modelling, this paper proposes a method for reconstructing a parametric CAD assembly from multi-view images. SAM 2 extracts the main body of the product from each view, and TRELLIS.2 uses this data to generate a mesh with material attributes. After axis alignment, the program segments components, identifies materials, and extracts exposed contours, then obtains key dimensions through dimensional calibration. The image reconstruction results are used to construct the exposed geometry, while design drawings provide hidden features and assembly constraints. Validation results for 16 insulators across 4 categories show that the macro-average F1 scores for surface category classification and component segmentation were 0.924 and 0.923, respectively; among the 48 major components, 44 were correctly identified by material, yielding an accuracy rate of 91.7%. The average absolute percentage error for independent length parameters was 2.32%. All 16 CAD assembly models were successfully re-imported into SolidWorks and meshed in COMSOL, with 15 of them passing the comprehensive delivery inspection. When product family templates and material reference libraries were reused, the average processing time per sample, measured from the input of 8 valid view images, was 134.6 s. The output STEP assembly models are linked to parameter sources, material labels, and inspection statuses, forming digital delivery units ready for use in engineering software. The method is applicable to similar equipment with approximately axisymmetric geometry, dimensional references, and corresponding design drawings.

Authors

Institutions

Publication Details

Journal
Applied System Innovation
Published
2026-11-12
DOI
https://doi.org/10.3390/asi9110239
Primary Topic
Manufacturing Process and Optimization
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Image-to-CAD Reconstruction for the Digital Delivery of Power Equipment: Validation Across Four Insulator Types

Zhiwei Rao, Hao Luo
Applied System Innovation
Manufacturing Process and Optimization
article

Image-to-CAD Reconstruction for the Digital Delivery of Power Equipment: Validation Across Four Insulator Types

Zhiwei Rao, Hao Luo
article en

Abstract

Digital delivery of power equipment requires integrating design information and the physical geometry of individual manufactured units into a single computer-aided design (CAD) model. To address the time-consuming nature of unit-by-unit measurement and manual modelling, this paper proposes a method for reconstructing a parametric CAD assembly from multi-view images. SAM 2 extracts the main body of the product from each view, and TRELLIS.2 uses this data to generate a mesh with material attributes. After axis alignment, the program segments components, identifies materials, and extracts exposed contours, then obtains key dimensions through dimensional calibration. The image reconstruction results are used to construct the exposed geometry, while design drawings provide hidden features and assembly constraints. Validation results for 16 insulators across 4 categories show that the macro-average F1 scores for surface category classification and component segmentation were 0.924 and 0.923, respectively; among the 48 major components, 44 were correctly identified by material, yielding an accuracy rate of 91.7%. The average absolute percentage error for independent length parameters was 2.32%. All 16 CAD assembly models were successfully re-imported into SolidWorks and meshed in COMSOL, with 15 of them passing the comprehensive delivery inspection. When product family templates and material reference libraries were reused, the average processing time per sample, measured from the input of 8 valid view images, was 134.6 s. The output STEP assembly models are linked to parameter sources, material labels, and inspection statuses, forming digital delivery units ready for use in engineering software. The method is applicable to similar equipment with approximately axisymmetric geometry, dimensional references, and corresponding design drawings.

Applied System InnovationVol. 9(11)
Chongqing University (CN)
National Natural Science Foundation of China
Industry, innovation and infrastructure
Manufacturing Process and Optimization
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon