Aiming at the problems of difficult springback control and poor forming accuracy during the CNC bending of aero-engine pipes, the CNC bending process experiments of 0Cr18Ni9 and 1Cr18Ni9Ti stainless steel pipes were carried out. The springback data of different pipe diameters, wall thicknesses, relative bending radius and bending angles were obtained. A springback prediction model of pipes during CNC bending was established through machine learning and genetic algorithm. After genetic algorithm optimization, the prediction error distribution of the model was reduced from [–0.741°, 0.771°] to [–0.310°, 0.314°]. On this basis, an intelligent prediction and compensation system for springback of pipes during bending was developed. It was used for the compensation on process parameters and springback control of typical full-size aero-engine pipes during CNC bending. After testing, it was found that the maximum angular deviation of the full-size aero-engine pipes compensated by the system was 0.358°, which fully met the actual production accuracy requirements, and significantly improved the CNC bending accuracy and production efficiency of the aero-engine pipes.