Aiming at the important item of safety maintenance for the manual daily maintenance of rolling stock and rolling stock, the multi-dimensional vision acquisition scheme is used to carry out real-time high-definition imaging of key components and systems of trains passing through the inspection area, including the walking part, traction, flow receiving, brake air supply, doors and windows, air conditioning, etc., and realize automatic identification and alarm of key train components abnormalities through artificial intelligence technologies such as computer vision, deep learning and data-driven hierarchical warning. The aim is to improve operational efficiency, reduce the amount of manual maintenance and lower O&M costs.