Physics-informed machine learning with domain decomposition and global dynamics for three-dimensional intersecting flows