Fast and accurate active alignment of camera lenses with physics-informed deep learning

Area of Science:

• Optical Engineering
• Machine Learning
• Manufacturing Technology

Background:

• Complex optical systems demand precise alignment for optimal performance.
• Active alignment (AA) uses real-time feedback but faces speed/accuracy challenges in mass production.
• Existing AA methods struggle with the demands of large-scale manufacturing, like smartphone lens production.

Purpose of the Study:

• To develop a novel active alignment pipeline leveraging physics-informed deep learning.
• To enhance the speed and accuracy of optical alignment in large-scale manufacturing.
• To address the limitations of current AA techniques in high-volume production environments.

Main Methods:

• Proposed a two-component pipeline: a physics-informed tolerance estimation neural network (TolNet) and an optical optimization module.
• TolNet estimates tolerances from point spread functions (PSFs) using a hybrid data-driven and physics-driven loss strategy.
• The optical optimization module determines adjustment parameters for AA.

Main Results:

• TolNet achieved exceptional speed, completing tolerance estimation in under 0.01 seconds.
• The optical optimization module required less than 3 seconds for adjustment parameter determination.
• The proposed method demonstrated improved efficiency and accuracy in validation experiments.

Conclusions:

• The physics-informed deep learning pipeline offers a promising solution for efficient and accurate active alignment.
• This approach enhances AA performance in large-scale manufacturing settings.
• The method effectively balances speed and accuracy for complex optical system production.