Open Access

ARTICLE

MMNet: Integration Multi-Attention and Multi-Strategy Network for Feature Recognition

Shuai Ma¹, Xiang Fang^1,2,*, Liya Han¹

1 Chengdu Aircraft Industrial (Group) Co., Ltd., Chengdu, China
2 National College for Excellent Engineers, Beihang University, Beijing, China

* Corresponding Author: Xiang Fang. Email:

(This article belongs to the Special Issue: Data-Driven and Physics-Informed Machine Learning for Digital Twin, Surrogate Modeling, and Model Discovery, with An Emphasis on Industrial Applications)

Computer Modeling in Engineering & Sciences 2026, 147(2), 39 https://doi.org/10.32604/cmes.2026.078073

Received 23 December 2025; Accepted 03 April 2026; Issue published 27 May 2026

Abstract

Automated feature recognition (AFR) plays an important role in automated measurement path planning and metrological data processing in the manufacturing industry. Existing AFR methods face critical limitations, such as the loss of geometric-topological fidelity during Computer-aided design (CAD) model conversion and inadequate instance segmentation for dimensional metrology. To address these challenges, we propose an integrated multi-attention and multi-strategy network (MMNet) for feature recognition, which mainly includes the multi-attention geometric and attribute fusion module (MGAM) and the multi-strategy semantic and instance segmentation module (MSIM). Specifically, MGAM employs multi-attention mechanisms to synergize local geometric features with global attributes of the boundary representation (B-rep) to enhance recognition precision. Subsequently, MSIM integrates a graph neural network for face-level semantic segmentation with rule-based instance clustering to achieve robust feature recognition. Experimental results demonstrate that our proposed method outperforms other state-of-the-art methods on MFCAD, MFCAD++, and our real-world datasets.

---

Keywords

---