When　mechanical　products　work　in　complex　environments,　it　is　imperative　to　build　an　optimal　maintenance　strategy,　based　on　accurate　positioning　of　fault　locations　and　prediction　of　fault　conditions.　Based　on　digital　twinning　technology,　this　paper　proposes　a　＂super-network-warning　features＇＇　fault　prediction　and　maintenance　method.　According　to　the　digital　twin　five-dimensional　structure,　a　three-layer　super-network　model　is　constructed,　providing　a　quantitative　research　for　data　among　heterogeneous　subjects　in　digital　twinning.　Early-warning-features　in　the　physical　layer,　virtual　layer　and　service　layer　are　selected　as　input　parameters　of　the　fault　prediction　model　to　accurately　predict　the　cause　of　the　fault.　Then,　using　the　simulation　and　optimization　functions　of　the　virtual　model　in　digital　twinning,　a　real-time　maintenance　strategy　is　formulated　for　the　causes　of　the　fault.　It　supplements　the　missing　link　between　fault　prediction　and　maintenance.　Taking　an　aero-engine　bearing　as　an　example,　this　method　is　compared　with　a　traditional　method.　The　results　show　that　the　model　prediction　error　of　this　method　is　better　than　the　traditional　method.