Imaging quality prediction and preload optimization for a precision optical system considering assembly errors

Abstract

With the gradual enhancement in the accuracy of machining and the consistency of quality in precision optical components, the assembly process has emerged as a critical phase influencing the imaging quality of precision optical systems. The ability to accurately and swiftly predict the imaging quality of these systems under various assembly errors, along with optimization of processing, can facilitate the assembly of optical systems. In this research, an integrated optomechanical simulation approach was introduced, taking into account various assembly errors. This approach involved simulating the tightening process of bolts within the optical system, and the resulting mirror deformation due to bolt preloading was derived and fitted using Zernike polynomials. Furthermore, Zeman software was utilized to model the imaging quality of the optical system, factoring in both mirror deformation and posture deviations. The effects of different assembly errors on the energy concentration of the optical system were systematically examined. A dataset comprising preloads, posture deviations, and their corresponding energy concentrations was created. Ultimately, a surrogate model incorporating a combined MLP-XGBoost neural network was developed and trained on this dataset. The model's superiority and reliability regarding prediction accuracy were confirmed through comparisons with MLP and XGBoost models, as well as traditional regression models (BP and SVR). Additionally, the stochastic gradient descent method was applied to optimize the preload magnitudes under various posture deviations of the primary and secondary mirrors. The findings indicated an average improvement of 11.34% in energy concentration compared to the values prior to optimization. This method significantly enhances the assembly efficiency and precision of the dual-reflector optical system.