AI 中文总结
本文针对AtLAST主动射电望远镜，提出算法框架优化传感器布置，结合FEA与POD基，用50个传感器可将其表面重建残差控制在2.7μm rms以下，满足精度要求。
AI 中文摘要
下一代高频射电望远镜需要主表面精度，仅靠被动结构无法可靠实现。阿塔卡马大口径亚毫米波望远镜（AtLAST）是一个口径50米的单碟概念，工作频率最高约1太赫兹（THz），要求整个口径范围内的表面精度均方根（rms）约20微米（μm）。对于受重力、热和风致变形影响的纯被动反射器，这一精度实际无法达到。因此，闭环主动准直和表面控制必不可少，而这需要实时获取整个口径范围内的变形场。然而，以所需分辨率直接测量整个表面的变形几乎不可行；因此，当前AtLAST的概念开发计划从有限的离散传感器位置重建表面。本文提出一种算法框架，用于优化这些传感器的数量和布置，以最大化变形场的可重建性。有限元分析（FEA）载荷工况（重力、热、风）定义了变形空间，从中导出数据驱动的本征正交分解（POD）基；随后通过贪心优化算法选择传感器位置，并通过留一法交叉验证进行评估。将该方法应用于AtLAST的支撑结构（BUS）的FEA变形，假设传感器噪声为5μm rms，使用50个传感器可将所有载荷工况的重建残差（BUS表面误差）控制在2.7μm rms以下（最坏情况）。
英文摘要
Next-generation high-frequency radio telescopes require primary-surface accuracies that passive structures alone cannot reliably achieve. The Atacama Large Aperture Submillimeter Telescope (AtLAST), a 50 m single-dish concept operating up to $\approx$ 1 THz, imposes a $\approx$ 20 $μ$m rms surface-accuracy requirement across its full aperture. This is practically unattainable for a purely passive reflector subject to gravitational, thermal, and wind-induced deformation. Closed-loop active collimation and surface control are therefore imperative, which in turn requires the deformation field to be known across the full aperture in real time. Measuring it directly at the necessary resolution across the complete surface is, however, hardly feasible; instead, the current AtLAST concept development foresees reconstructing the surface from a limited set of discrete sensor positions. An algorithmic framework is presented that optimizes the number and placement of these sensors to maximize the reconstructability of the deformation field. Finite-element analysis (FEA) load cases (gravity, thermal, wind) define the deformation space, from which a data-driven Proper Orthogonal Decomposition (POD) basis is derived; sensor positions are then chosen by a greedy optimization algorithm and then assessed via leave-one-out cross-validation. Applied to FEA deformations of AtLAST's Back-Up Structure (BUS), and assuming a sensor noise of 5 $μ$m rms, the method reconstructs all load cases with 50 sensors to below 2.7 $μ$m rms (worst case) residual (BUS-) surface error.
CommentsSubmitted as an SPIE proceedings manuscript for the 2026 SPIE Astronomical Telescopes + Instrumentation Conference
