Leveraging graph representations as a natural learner for monitoring and twinning, with applications to railway, wind energy, and bridge asset