CoPR-DTO: Collaboratively driven and periodically recalibrated dynamic topology optimization

TY - GEN

T1 - CoPR-DTO

T2 - 2025 International Conference on Pattern Recognition and Image Analysis, PRIA 2025

AU - Li, Xia

AU - Zhang, Kai

AU - Liu, Guangyan

N1 - Publisher Copyright: © 2026 SPIE.

PY - 2026/4/15

Y1 - 2026/4/15

N2 - Topology optimization seeks optimal material distribution in a design domain to meet constraints and maximize performance. While static load optimization is well-developed, real-world structures often face dynamic loads, making dynamic characteristics critical. Traditional dynamic topology optimization methods are computationally expensive due to iterative dynamic analysis, limiting their applicability to complex problems. This paper proposes a collaboratively driven dynamic topology optimization framework with periodic recalibration. It integrates two neural networks: a topology generation network mapping coordinates to element densities, and a U-Netbased dynamic response prediction network for fast displacement field prediction. A periodic recalibration strategy maintains model accuracy: initial high-fidelity data trains the prediction network, which then drives rapid optimization iterations, with periodic reversion to finite element solvers for data update and model recalibration. Dual loss functions and gradient separation ensure independent network optimization. This framework balances computational efficiency and physical consistency, offering an effective solution for complex dynamic topology optimization.

AB - Topology optimization seeks optimal material distribution in a design domain to meet constraints and maximize performance. While static load optimization is well-developed, real-world structures often face dynamic loads, making dynamic characteristics critical. Traditional dynamic topology optimization methods are computationally expensive due to iterative dynamic analysis, limiting their applicability to complex problems. This paper proposes a collaboratively driven dynamic topology optimization framework with periodic recalibration. It integrates two neural networks: a topology generation network mapping coordinates to element densities, and a U-Netbased dynamic response prediction network for fast displacement field prediction. A periodic recalibration strategy maintains model accuracy: initial high-fidelity data trains the prediction network, which then drives rapid optimization iterations, with periodic reversion to finite element solvers for data update and model recalibration. Dual loss functions and gradient separation ensure independent network optimization. This framework balances computational efficiency and physical consistency, offering an effective solution for complex dynamic topology optimization.

KW - Collaborative driving

KW - Dynamic topology optimization

KW - Neural network

KW - Periodic recalibration

KW - Surrogate model

UR - https://www.scopus.com/pages/publications/105039430951

U2 - 10.1117/12.3111177

DO - 10.1117/12.3111177

M3 - Conference contribution

AN - SCOPUS:105039430951

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - International Conference on Pattern Recognition and Image Analysis, PRIA 2025

A2 - Ma, Jixin

A2 - Fournier-Viger, Philippe

A2 - Zheng, Qian

A2 - Jain, Deepak Kumar

PB - SPIE

Y2 - 26 December 2025 through 28 December 2025

ER -