Towards compact AI models for efficient machining feature recognition

Tools

Gkrispanis, Konstantinos and Nousias, Stavros and Borrmann, André; Moreno-Rangel, Alejandro and Kumar, Bimal, eds. (2025) Towards compact AI models for efficient machining feature recognition. In: EG-ICE 2025. University of Strathclyde Publishing, GBR, pp. 298-305. ISBN 9781914241826 (https://doi.org/10.17868/strath.00093276)

[thumbnail of Gkrispanis-etal-EG-ICE-2025-Towards-compact-AI-models-for-efficient-machining]
Preview
 Text. Filename: Gkrispanis-etal-EG-ICE-2025-Towards-compact-AI-models-for-efficient-machining.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (1MB)| Preview

Abstract

Machining feature recognition is the first step in the automation of the design and production pipeline. Currently, this process relies on manual annotation by human experts, which is time-consuming and prone to errors. Computer Numerical Control (CNC) machines are automated tools that use pre-programmed computer software to control machining processes with high precision and efficiency. Enhancing CNC machines with an AI-based approach for the recognition of machining features in the CAD (Computer Aided Design) input models eliminates the need for manual annotation and enables seamless integration of design and production workflows for optimized machining strategies. CNC controllers often operate in resource-constrained environments with limited computational capabilities. Therefore, there is a pressing demand for machining feature recognition models that can operate efficiently across these devices. In recent years, network pruning algorithms have gained significant attention from researchers due to the growing size and complexity of deep learning models, which often require considerable computational resources for training and development. Network pruning is a technique that reduces the size of deep learning models by removing unnecessary weights or entire structures (e.g., filters, channels). Despite their growing adoption in other domains, pruning strategies have not been explored in machining-specific AI models. In this paper, we evaluate four different scoring criteria combined with the Soft Pruning iterative procedure on BRepNet, a machining feature recognition model. Our experiments demonstrate that pruning not only preserves performance but can also lead to slight accuracy improvements for small pruning rates. Remarkably, when removing 90% of the model’s parameters, one pruning criterion results in only 2% loss in accuracy. These findings highlight the potential of pruning as a practical approach to developing efficient and compact AI models for deployment in manufacturing and robotized construction environments.

  • Item type:Book Section
    ID code:93276
    Dates:
    Date
    Event
    1 July 2025
    Published
    9 June 2025
    Accepted
    Subjects:Fine Arts > Architecture
    Department:Faculty of Engineering > Architecture
    Depositing user: Pure Administrator
    Date deposited:27 Jun 2025 10:57
    Last modified:01 Apr 2026 15:06
    URI:https://strathprints.strath.ac.uk/id/eprint/93276
CORE (COnnecting REpositories)