FEM-driven machine learning approach for characterizing stress magnitude, peak temperature and weld zone deformation in ultrasonic welding of metallic multilayers: application to battery cells