Validation of neural networks model based on beam dynamics simulation for an automated control in high-intensity proton injector

We have studied the feasibility of automatic and fast control of the low energy beam transport section in the newly developed radio-frequency quadrupole at KOMAC (Korea Multipurpose Accelerator Complex) by combining a simple neural network model with a beam profile monitor. Extensive beam dynamics simulations on the proton injector with varying beam transport parameters are performed to generate the training set for the machine learning and to obtain optimization model for the injector. These datasets are well-trained and show good ability to estimate the beam parameters at position under consideration. These results can be a steppingstone to develop an auto-tuning or feedback control system based on artificial intelligence for a high-intensity accelerator. This paper presents the calculation conditions and the training process in detail, as well as the cross-validation of the trained neural network model by using the results obtained by beam dynamics code.