An assisted decision-making tool for synchrotron beamline alignment based on neural networks

Abstract

To achieve an excellent focus quality, the parameters of optical elements (OEs) are, in most of the synchrotron beamlines, manually adjusted. This procedure is not only time-consuming and experience-dependent but also extremely complex when various experimental requirements are involved. Responding to this challenge, we propose a new beamline alignment tool based on neural network-assisted design. This method can predict the parameters of OEs, according to experimental requirements. Specifically, the artificial neural network (ANN) training set is generated, based on SHADOW3 and Synchrotron Radiation Workshop (SRW) in OASYS. Then, the magnification factor (M) of the focusing lens and the position (P) of the secondary source is predicted, using the aforesaid tool. Finally, the parameters are verified by substituting back to the OASYS. The results show that learned NNs can predict the main parameters of the OEs with high accuracy (above 97%). Then bring the parameters above back to the OASYS software to obtain the re-tracing results. Furthermore, the final focused quality at the sample point satisfies the experimental design indicators. Experimental design indicators are flux, full width at half-maximum (FWHM) at the sampling point and transmission efficiency. Compared to other methods, this is a successful exploration of the ANN in the field of synchrotron beamline alignment, and it is an important guide for the design of beamlines alignment.