Beamline Steering Using Deep Learning Models

TL;DR

The paper tackles beamline steering for light sources, focusing on LTU where frequent recalibration undermines traditional SVD-based methods due to nonlinear effects at higher amplitudes. It compares forward and inverse deep learning mappings using archival EPICS data and PyTAO simulations, finding a high-accuracy forward model (8-layer FFNN, RMSE ~$3.4\times10^{-3}$) but limited generalization for the inverse model to simulated data, and overall DL performance lagging behind SVD. The results demonstrate the potential of DL-based steering on real data, while highlighting challenges from covariate shifts and ill-posed inverse mappings, underscoring the need for robustness improvements before deployment. The work suggests forward DL models can reduce operator burden and enable faster, adaptive beam steering, albeit with careful attention to data representativeness and system nonlinearities.

Abstract

Beam steering involves the calibration of the angle and position at which a particle accelerator's electron beam is incident upon the x-ray target with respect to the rotation axis of the collimator. Beam Steering is an essential task for light sources. The Linac To Undulator is very difficult to steer and aim due to the changes of each use of the accelerator there must be re-calibration of magnets. However with each use of the Beamline its current method of steering runs into issues when faced with calibrating angles and positions. Human operators spend a substantial amount of time and resources on the task. We developed multiple different feed-forward-neural networks with varying hyper-parameters, inputs, and outputs, seeking to compare their performance. Specifically, our smaller models with 33 inputs and 13 outputs outperformed the larger models with 73 inputs and 50 outputs. We propose the following explanations for this lack of performance in larger models. First, a lack of training time and computational power limited the ability of our models to mature. Given more time, our models would outperform SVD. Second, when the input size of the model increases the noise increases as well. In this case more inputs corresponded to a greater length upon the LINAC accelerator. Less specific and larger models that seek to make more predictions will inherently perform worse than SVD.

Figures (3)

• Figure 1: The neural network model developed for the inverse modeling paradigm: Neural Network Model Architecture
• Figure 2: Training history for the neural network model developed for the inverse modeling paradigm: Root Mean Squared Error For Test and Training data
• Figure 3: Inverse model predictions on archival data and simulation data