Tandem neural network-based controller for x-ray bimorph mirrors

作者：Runyu Zhang, Luca Rebuffi, Christopher Egly, Xianbo Shi, Lahsen Assoufid · 发表于：Optics Letters · 年份：2025 · DOI：10.1364/ol.559583 · 被引用次数：2 · 研究领域：Adaptive optics and wavefront sensing、Optical Systems and Laser Technology、Geophysics and Sensor Technology

Nanometer-scale shape control of x-ray mirrors is crucial for coherent x-ray beam experiments at low-emittance synchrotron beamline instruments. Piezoelectric bimorph mirrors offer adaptive control but are hindered by nonlinearities such as cross talk, creep, and hysteresis. To overcome these limitations, we present a novel feedback-free control solution, inspired by the proportional–integral–derivative (PID) scheme, driven by tandem neural networks (TNNs). Using task-specific datasets, the TNN-based system predicts actuator voltages with greater speed, accuracy, and stability than a single NN-based model. This approach is ideal for real-time applications, such as adapting beam focus to dynamic sample sizes while maintaining precise wavefront quality. Our findings highlight the potential of artificial intelligence in rapidly optimizing adaptive optics and managing nonlinear control systems.