Surface shape prediction of optical lenses using neural networks

Reflective optical system is critical component in optical devices used for laser processing, optical imaging, detection, etc., with extensive application and extremely high precision requirements. The precision of reflective optical mirrors is influenced by multiple factors and this paper focuses on assembly errors. By establishing a finite element mechanical model for the optical lens assembly process and analyzing the deformed mirror surfaces in the results, the influence patterns of assembly-induced errors are investigated and Zernike coefficients are obtained. By randomly generating combinations of 3 forces and obtaining the corresponding Zernike coefficients, a database consisting of 2000 data points is established. Based on this database, MLP neural network is established to predict the Zernike coefficients of surface shape induced by assembly force, so as to predict the surface shape of the mirror. Finally, the validity of the neural network is verified through residual analysis and the influence sensitivity of bolt loads on Zernike coefficients is examined. The model proposed in this paper has been proved to be effective and precise. Moreover, compared with traditional analysis methods, it can achieve rapid and real-time prediction of the surface shape.