Real-time machine-learning-driven control system of a deformable mirror for achieving aberration-free X-ray...

Area of Science:

• Optics
• Machine Learning
• X-ray Science

Background:

• Adaptive optics are crucial for maintaining coherent X-ray wavefronts at advanced light sources.
• Existing control methods for deformable mirrors often lack the precision and speed required for dynamic beamline conditions.

Purpose of the Study:

• To develop and validate a neural-network-based machine learning model for controlling bimorph adaptive mirrors.
• To achieve and preserve aberration-free coherent X-ray wavefronts at synchrotron radiation and free electron laser beamlines.

Main Methods:

• A neural-network machine learning model was trained using directly measured actuator responses from a bimorph mirror.
• A real-time single-shot wavefront sensor employing a coded mask and wavelet-transform analysis was utilized for training.
• The system was tested on a bimorph deformable mirror at the Advanced Photon Source 28-ID IDEA beamline.

Main Results:

• The model achieved a response time of a few seconds.
• Sub-wavelength accuracy in maintaining desired wavefront shapes (e.g., spherical) was demonstrated at 20 keV X-ray energy.
• Performance significantly surpassed that of linear control models.

Conclusions:

• The developed neural-network controller effectively manages adaptive mirrors for high-quality X-ray wavefronts.
• The system's adaptability allows for application to various mirror types and actuators.
• This approach offers a significant improvement for coherent X-ray applications at light source facilities.