基于机器学习的大面积拼接镜倾斜误差探测

Tip-tilt error detection of large segmented mirror based on machine learning

为了实现对拼接镜整体幅面的共相位误差的快速检测,通过Zemax建立相位测量装置,数值模拟拼接镜的倾斜误差检测过程。使用基于主成分分析(PCA)、偏最小二乘回归(PLSR)的机器学习算法,替代传统的相位重建方法,从探测面强度分布图中提取倾斜误差。预测结果表明,在单元情况下对12个样本的倾斜角进行预测,倾斜角预测值与真实值间的均方根误差(RMSE)约为0.00029;在多元情况下,倾角的RMSE均维持在0.0003以下。可见,在两种情况下,倾角的RMSE参量值均小于倾斜步长。因此,利用机器学习算法可以实现对倾角步长为0.0005°的倾斜误差的预测,与相位差波前检测等传统方法相比,该方法能大幅提高预测速度,明显降低传统波前重建算法复杂度。

In order to realize the rapid detection of the co-phasing error of the overall surface of the segmented mirror,a phase measurement device was established by Zemax to simulate the tilt error detection process of the segmented mirror.Instead of using conventional phase reconstruction method,machine learning based on the combination of the Principal Component Analysis(PCA)and the Partial Least Squares Regression(PLSR)algorithms is investigated to extract the tip-tilt error from the intensity pattern.The prediction results show that when single segment is tilted,the tilt angle of the 12 samples is predicted,and the Root Mean Square Error(RMSE)between the predicted value of the tilt angle and the true value is about 0.00029;When multiple segments are tilted.the RMSE of the tilt angle can be maintained below 0.0003.It can be seen that in both cases,the RMSE value of the tilt angle is less than the inclination step.Therefore,machine learning algorithms can be used to predict tilt errors with a step size of 0.0005°.Compared with traditional methods,such as the phase diversity wavefront sensing technology,this method can greatly improve the prediction speed and reduce the complexity of the traditional wavefront reconstruction algorithm.