Abstract: Aiming at the problems of low accuracy and long time consumption of straightening stroke prediction of shaft straightening machines, this paper proposes an improved fuzzy neural network model. The system framework of the prediction model is a combination of the fuzzy system and the neural network, and a connection layer is designed in the network structure, which can give feedback to the historical data of the straightening stroke and enhance the network data processing ability. Relevant factors affecting the straightening stroke are taken as reference indexes, successful real-time straightening data are taken as model input, and the straightening stroke is taken as model output. Compared with the traditional prediction methods, the experimental results show that the relative error between the actual value and the predicted value of the improved fuzzy neural network is 1.65%, achieving an improvement in the prediction accuracy of the straightening stroke and the straightening efficiency.