用代理模型实现了一个新的失调量求解方法,可在较大失调范围内进行更为准确的求解。首先分析了基于灵敏度矩阵方法求解失调量的原理和局限性,提出了使用代理模型求解失调量的设想。在代理模型方法中,设定光学元件各个自由度对应的失调范围,对各自由度失调量进行大量的蒙特卡洛仿真得到充足的样本模型。然后从样本模型中提取失调量和残余像差数据,其中残余像差数据用Zernike系数表示。对比三种不同的代理模型,选择用人工神经网络表示失调量与残余像差之间的非线性映射关系或统计关系。最后用一个离轴三反射系统模拟装调的例子,说明了如何建立代理模型,并根据当前系统的状态进行调整,表明了使用代理模型进行辅助装调的实用性和可靠性。
As the supplement for the current computer aided alignment method for the advanced optical system, surrogate model was proposed to represent the intricate relationship between the misalignment parameters and the aberration coefficients, so a new solution method of the misalignments could be carried out. At the beginning, the principle of the sensitivity matrix method was analyzed to show its limitation, and then the surrogate model was introduced as a new tool. Comparison of various surrogate models, the artificial neural network was chosen to represent the obscure relationship between the misalignment parameters and the wavefront error, while these data were collected by Monte Carol simulation in Zemax. In the end, an off axis three mirrors astigmatism system was taken as an example to illustrate the process of building the surrogate model and calculating the misalignment. All above procedures show the practicability and reliability of the surrogate method for the optical alignment, and it is more robust and quick than the conventional sensitivity marx method.