Automatic alignment systems are becoming increasingly crucial for complex beamlines such as those of the APS upgrade project. Manual methods are no longer sufficient to achieve optimal configurations, and as such, at the APS we are developing AI-powered adaptive optics control and integrating it through machine learning. This technology has the potential to save time and maintain coherent beam wavefronts, making it a valuable addition to the APS’s operations.
Our approach involves a neural network-driven controller that works in conjunction with a high-speed X-ray wavefront sensor. The method has been proven to improve the electrodes of a bimorph mirror to achieve sub-wavelength accuracy in shaping wavefronts in just a few seconds, especially for spherical wavefronts.
In addition, a Bayesian optimization system is being developed to auto-align nanofocusing KB mirrors. This system will create a digital replica of existing beamlines, calibrated using actual experimental data. By conducting virtual experiments with this digital twin, the optimization process can be enhanced to achieve desired focal spot sizes and positions in a more efficient manner.
Our preliminary testing of the system on a real optical setup at the APS has shown the viability of this approach including optics alignment and beam wavefront shaping. These advancements are helping us progress steadily towards the objective of fully automatic beamline operation and optimization, opening up new research possibilities.