Optimization of mirror support structures based on CNN-LSTM

With the advancement of modern optical technology, the design requirements for large-aperture mirror support structures have become increasingly stringent. These structures must maintain mirror precision and structural rigidity under complex conditions such as static gravitational loads and low-frequency vibrations. However, traditional design methods often struggle to efficiently address the nonlinear coupling between support structures and performance, while also incurring high computational costs. To address this, this paper proposes a surrogate model combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) for the optimized design of large-aperture mirror support structures. The model utilizes CNN to extract spatial features from input design parameters and employs LSTM to capture dependencies between parameters, accurately reflecting the complex nonlinear mapping relationship between design variables and structural response. Results demonstrate that the model efficiently captures the relationship between structural performance and design parameters, achieving low prediction errors and significantly enhancing optimization efficiency. During the optimization process, by adjusting key parameters of the support structure, the final optimized results achieved a mirror surface accuracy of 8.86 nm, a dynamic frequency of 64.584 Hz, and a maximum stress of 0.462 MPa. This method provides a novel solution for the efficient optimization design of large-aperture mirror support structures.