The intelligent assembly technology of the industrial production line was a key means to improve the work efficiency of the production line and ensure the uniformity of the products. Increasingly complex industrial product structures have higher requirements for intelligent assembly control accuracy. Among them, feedback correction technology based on real-time online measurement has become a research hotspot. In order to realize the flexible measurement of the intelligent assembly process of industrial production lines, a feedback assembly control method based on optical fiber sensing was proposed. The system uses multiple FBG sensors to obtain the real-time strain of the assembly structure, and the temperature strain decoupling was completed by different packaged FBG sensors. Finally, the conversion of the degree of position deviation was completed by the way of neural network learning, and real-time correction of the assembly position was realized. Matlab was used to simulate the screw structure assembly process, and the stress distribution of screws with different lengths and force levels was simulated. The results show that the longer the screw length, the greater the strain under the same force. And the main strain of the screw is concentrated on the edge of the screw hole. In the experiment, the cantilever beam strain calibration experiment is used to obtain the fitting curves of different package states, the temperature calibration experiment is used to eliminate the temperature drift effect in the strain test data, and the position shift experiment is used to obtain the correction parameters. The results show that for a screw with a length of 15.0 mm, the average resolution of the system is about 0.012 3 mm/N, and for a screw with a length of 20.0 mm, the average resolution of the system is about 0.022 1 mm/N, and it has good linearity.