Opto-thermal deformation fitting method based on a neural network and a transfer learning.

Area of Science:

• Optics
• Mechanical Engineering
• Thermal Analysis
• Artificial Intelligence

Background:

• Accurate thermal deformation fitting is crucial for reliable optical-mechanical-thermal integrated analysis.
• Traditional numerical methods struggle with fitting accuracy, efficiency, and high-order Zernike polynomials.
• Existing methods face limitations in handling complex thermal deformation data.

Purpose of the Study:

• To develop an innovative opto-thermal deformation fitting method using neural networks and transfer learning.
• To overcome the limitations of traditional numerical methods in fitting optical surface thermal deformations.
• To improve the accuracy and efficiency of thermal deformation fitting for optical systems.

Main Methods:

• A one-dimensional convolutional neural network (1D-CNN) was trained with Zernike polynomials as input and optical surface sag change as output.
• Transfer learning was employed to efficiently fit thermal deformations across different temperatures without retraining.
• The method was validated using thermal analysis on an aerial camera's main mirror.

Main Results:

• The 1D-CNN model achieved a determination coefficient greater than 99.9% in regression analysis.
• Predicted Zernike coefficient distributions from 1D-CNN and transfer learning showed high consistency.
• The proposed method demonstrated significant reductions in peak-valley, root mean square, and mean relative errors compared to the least squares method.

Conclusions:

• The proposed neural network and transfer learning method significantly enhances the accuracy and efficiency of optical surface thermal deformation fitting.
• This advancement leads to more reliable optical-mechanical-thermal integrated analysis.
• The method offers a robust solution for handling complex thermal deformation challenges in optical systems.