The present invention relates to the field of
mechanical engineering, and specifically to a method for predicting the
imaging quality of an optical
system based on
assembly errors. First, an accurate finite
element model of the optical
system is constructed, the bolt tightening process is simulated, the mirror deformation under the action of
assembly stress is obtained, and the mirror
surface shape is fitted using
Zernike polynomials. Using this as an intermediate medium,
Zemax is used to simulate the imaging of the optical
system under the conditions of mirror deformation and
assembly posture, and the influence of various assembly errors on the
imaging quality of the optical system is explored. Then, an assembly
data set is constructed through an integrated
simulation method, an MLP-XGBoost combined neural
network agent model is established, and model training is performed. Compared with single MLP, XGBoost models, and traditional regression models, the model has greater superiority and reliability in terms of prediction accuracy. Finally, the
stochastic gradient descent method is used to optimize the preload force under different primary and
secondary mirror assembly posture conditions. The energy concentration is increased by an average of 11.34% compared to before optimization, thereby improving the assembly efficiency and accuracy of the double-mirror optical system.