Mechanism-informed dynamic surrogate modeling for rapid thermal-aberration prediction in bonded optical systems
Related Experiment Video
Updated: Nov 1, 2026

Simulating the Mechanics of Lens Accommodation via a Manual Lens Stretcher
Published on: February 23, 2018
Yuxi Zhao1, Huanxiong Xia1,2,3,4, Xiumin Zhang1
1School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081, China.
Abstract:
Bonded optical systems exhibit thermal-history-dependent aberration evolution due to coupled thermoelastic deformation and viscoelastic relaxation, making rapid prediction under complex thermal loading a challenging engineering problem. Conventional thermo-structural-optical simulations are computationally expensive, while purely data-driven time-series models often exhibit limited generalization with insufficient training data. This study proposes a mechanism-informed dynamic surrogate model that explicitly incorporates thermoelastic and viscoelastic mechanisms for rapid thermal-aberration prediction. A thermoviscoelastic-structural-optical coupled simulation framework is first established to generate system-level Z5-Z9 aberration responses of a bonded two-mirror system under diverse thermal histories. The proposed model combines instantaneous thermoelastic effects with delayed viscoelastic relaxation to capture the dynamic evolution of thermal aberrations. A long short-term memory (LSTM) network is employed as a purely data-driven benchmark. For four unseen noncyclic thermal histories involving variations in temperature amplitude, heating/cooling rate, nonlinear temperature profiles, and holding duration, the proposed model achieves average normalized root mean square error (NRMSE) and normalized mean absolute error (NMAE) values of 0.72% and 0.53%, respectively. For two unseen multi-cycle thermal histories, the corresponding values remain as low as 2.20% and 1.76%. Training-set-size sensitivity analysis further demonstrates greater robustness of the proposed model than the LSTM benchmark as the training data are reduced. These results demonstrate that embedding thermoelastic and viscoelastic mechanisms improves prediction accuracy, robustness, and generalization under limited training data, providing an interpretable and computationally efficient approach for the design, analysis, and optimization of bonded optical systems.
