Researchers from the Shenguang II team at the Joint Laboratory for High Power Laser Physics, Shanghai Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, have developed a novel machine learning-based technique for crystal self-alignment in high-power laser facilities. Their research, titled "Crystal’s Self-Alignment for High Power Laser Facility Based on Machine Learning," was recently published in the IEEE Photonics Journal.
Online alignment of harmonic conversion crystals within high-power laser systems has long been a challenging and labor-intensive task. Traditional methods require operators to manually search for and adjust the position of the crystal's reflected spot, a process that is inefficient and struggles to guarantee precision. As laser power continues to increase, the demands for both the accuracy and speed of crystal alignment have intensified. Achieving rapid, automated, and high-precision crystal alignment has become a key technological bottleneck limiting the performance enhancement of high-power laser facilities.
To overcome this challenge, the research team employed a novel approach. They sampled the crystal alignment beam using grating diffraction and utilized a machine learning algorithm to automatically locate the reflected spot from the crystal's rear surface. Based on this location, the crystal position is then adjusted for optimal alignment. The team designed a system comprising two core modules: Rectangular Spiral Spot Scanning Module: Rapidly locates the initial position of the reflected spot. Automatic Spot Alignment Module (based on open-source M-LOOP algorithm): Employs Bayesian optimization with a Gaussian process probabilistic surrogate model to achieve precise alignment.
The hardware system integrates the crystal alignment optics, motors, a CCD camera, and a Raspberry Pi for control. Extensive experiments conducted on a large-scale laser facility demonstrated that this method can automatically search for and align the crystal's reflected spot in approximately 10 minutes, significantly enhancing alignment efficiency.
This research marks the first successful application of machine learning algorithms to the crystal alignment process within high-power laser facilities, realizing a fully automated workflow that eliminates dependence on manual operation. The highly efficient search and alignment algorithms drastically reduce alignment time.
This technology provides crucial support for the automated operation of high-power laser systems. It not only improves operational efficiency but also lays the groundwork for achieving higher-precision laser control. In the future, this method holds promise for broader application in other fields requiring precise optical alignment, potentially driving advancements in related technologies.
The research was supported by the Chinese Academy of Sciences Strategic Priority Research Program and the CAS Youth Innovation Promotion Association Project.
The paper's link:http://doi: 10.1109/JPHOT.2025.3578673
