Adaptive wavefront interferometry for large surface figure error utilizing convolutional neural networks

In order to solve the problems of large surface figure errors in large value, high spatial frequency and multi-aberration modes which are beyond the dynamic range of the interferometer, and the problems of slow iterative convergence and high non-convergence rate of existing adaptive compensation optimization algorithm, an adaptive wavefront interferometry utilizing convolutional neural network(CNN) for large surface figure error is proposed, based on the existing adaptive compensation interference detection methods. This paper first introduces the aberration regulation principle of spatial light modulator (SLM), and sets up a convolutional neural network. Then, SLM is controlled to generate Zernike aberrations with different coefficients. Combined with Zygo Verifire interferometer, the corresponding far-field light intensity is collected to compose a labeled data set to train CNN. Finally, a large surface figure error coefficient prediction experiment is carried out with the trained CNN, and the aberration compensation is performed according to the prediction coefficient to verify the effectiveness of the method. The experimental results show that the dense fringes can be transformed into resolvable fringes with this method, and the resolvable probability of full-diameter fringes after compensation is 66.7%. This method is able to greatly improve the performance of adaptive compensation detection, thereby meeting the demand for high dynamic range interference detection technology in the ultra-precision optical surface manufacturing process.