Machine learning-driven design and performance prediction of volume Bragg gratings for the Fabry–Perot cavity

The in situ sensing of parameters such as pressure and vibration in extreme environments poses a demand for high-temperature resistance, and Fabry–Perot (F-P) cavities based on quartz volume Bragg gratings (VBGs) are well-suited to this demand. The fabrication of quartz VBGs relies on femtosecond lasers, and the method of optimizing the performance of VBGs through trial and error requires considerable time investment. Machine learning technology, which performs tasks through statistical techniques and numerical algorithms, offers a new approach to addressing this challenge. This study proposed and constructed a machine learning-driven bidirectional intelligent prediction model for the relationships between parameters and performance of VBGs. Forward prediction from parameter to performance was implemented using the outlier labeling random forest (OLRF) model, which integrates the random forest algorithm with an outlier labeling optimization method. The model achieves R 2 values of 0.994 and 0.992 on the test set, respectively. Inverse prediction from performance to parameter was constructed using particle swarm optimization and genetic algorithms, with errors of less than 1% in reflectivity and 0.1 nm in FWHM. The OLRF–PSO model completes the inverse design of VBG micro-parameters in only 25 s. To demonstrate the practical feasibility, the designed VBG structures are applied to a Fabry–Perot cavity simulation. The results confirm that the generated grating parameters are well-suited for cavity integration, thereby providing data and technical support for the design optimization and application of F-P cavity-based optical sensors.