Inverse Surrogate Model of a Soft X-Ray Spectrometer using Domain Adaptation

TL;DR

This work develops an inverse surrogate model that maps detector images from a soft X-ray spectrometer to absolute spectrometer coordinates $x$, $y$, and $z$, enabling automated alignment. To overcome limited real data, the authors train on large-scale simulated data from RAYX and apply data augmentation plus adversarial domain adaptation via a gradient reversal layer to align synthetic and experimental domains. On 1,620 experimental images, augmentation plus domain adaptation improves $x$- and $y$-axis predictions, while $z$ predictions remain difficult due to weak image dependence; the approach demonstrates feasibility of robust, data-efficient automated beamline alignment. The method has practical impact by reducing alignment time during beamtime and enabling extension to other beamline components with minimal labelled real data.

Abstract

In this study, we present a method to create a robust inverse surrogate model for a soft X-ray spectrometer. During a beamtime at an electron storage ring, such as BESSY II, instrumentation and beamlines are required to be correctly aligned and calibrated for optimal experimental conditions. In order to automate these processes, machine learning methods can be developed and implemented, but in many cases these methods require the use of an inverse model which maps the output of the experiment, such as a detector image, to the parameters of the device. Due to limited experimental data, such models are often trained with simulated data, which creates the challenge of compensating for the inherent differences between simulation and experiment. In order to close this gap, we demonstrate the application of data augmentation and adversarial domain adaptation techniques, with which we can predict absolute coordinates for the automated alignment of our spectrometer. Bridging the simulation-experiment gap with minimal real-world data opens new avenues for automated experimentation using machine learning in scientific instrumentation.

Figures (5)

• Figure 1: Principle of an RZP, kubin_manganese_2019. In our spectrometer, the distance d is 90 mm, and the focal length f is 400 mm. The sample is a 4 mm-long wire composed of Cu$86$, Mn$12$, and Ni$2$. The CCD image shows the diffraction and focusing of Mn, along with the separation achieved for oxygen and 0th-order light.
• Figure 2: The left image shows the simulation, the middle image represents the experiment, and the right image depicts the simulation with augmentations applied.
• Figure 3: Structure of the adversarial domain adaptation network, showing the data paths for both synthetic and experimental data. The feature extractor is a convolutional network, the regressor and domain discriminator are fully-connected networks.
• Figure 4: Visualisations of the latent representations whilst preforming inference using experiment data.
• Figure 5: Inference results using 1,620 experimental images. True vs. predicted positions (x, y, and z) are shown, with rows representing no augmentation, data-specific augmentation, and augmentation with domain adaptation (top to bottom).