Speaker
Description
The complex electromagnetic and structural response of pulsed septum electromagnets is currently being studied, aiming to reveal insights in the prediction of their life cycles. The fault prediction remains a key challenge, requiring efforts in modelling and onsite instrumentation. Recently, a high-fidelity numerical model was developed, which allowed us to deepen our understanding of the system's behavior. The insight on critical stress/strain conditions are now allowing to better understand early fatigue observations in the coil conductor and required peripheric equipment. As a next step, a numerical parametric study has been conducted, enabling prediction of the mechanical and electromagnetic dynamic response under different operation conditions. A stringent experimental campaign has been carried out to validate these results. These tests required developing the test-setup including careful selection of non-contact and on-device instrumentation, as well as the data acquisition system. The setup has demonstrated the feasibility of potential implementation in the CERN accelerator complex. In parallel, significant efforts have been made to study the possibility of using Machine Learning (ML) models to reduce the time-consuming simulation process based on analytical solvers. Several custom loss functions reflecting physical constraints, as well as augmentation techniques, have been implemented to overcome the challenge of a strongly limited dataset size. The created ML pipelines and dedicated Graphical User Interface (GUI) are now enabling instantaneous response of the device’s model. The paper will summarize the efforts and project status with the long-term vision to develop a digital twin for septum magnets.
