Machine learning based image processing technology application in bunch longitudinal phase information extraction
Xingyi Xu, Yimei Zhou, Yongbin Leng
Abstract
We report on the application of machine learning (ML) methods to extract longitudinal phase information such as parameters of the synchrotron damping oscillation. Parameters of the synchrotron damping oscillation are important for the evaluation of machine status and bunch stability. It is of concern to extract these parameters with high-speed and high-precision. The previous methods, such as multiparameter nonlinear fitting and table look-up, are slower and easily fall into local optimal solutions. Our approach based on ML-image processing consists of training a virtual diagnostic to predict parameters using the beam position monitor (BPM) electrical signal data as inputs. We find that when the noise of data is large, our ML-model can still get better results than other methods, an important step toward on-line multiparameter extraction from multidimensional raw data.