Prediction of basic turnout rail straightening with a neural network

• Published in: Scientific reports Vol. 14; no. 1; pp. 23390 - 11
• Main Authors: Liu, Chunjiang, Dong, Zhikui, Hou, Xinyu, Qiao, Nanbing, Sun, Yongkuan, Qi, Xiangdong
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 08-10-2024; Nature Publishing Group; Nature Portfolio
• Subjects: 639/166/988; 639/301/1034; 639/705; Accuracy; Deformation; Humanities and Social Sciences; Load; Mathematical models; multidisciplinary; Neural networks; Predictions; Science; Science (multidisciplinary); Simulation; Trains
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-024-74652-6

The main factors affecting the rebound of a switch rail in the straightening process are analysed and used as the input data of a neural network. An experimental group is designed, finite element simulation is carried out, and the data results are used to train the designed neural network to obtain a model with high accuracy. Taking the U71Mn rail as an example, the parameters are selected according to a certain interval, the test dataset is obtained via finite element simulation, and the selected parameters are predicted by the neural network. The error between the detection prediction data and the simulation data is small, and the prediction results are divided to obtain the parameter range that meets the requirements. This method provides a reference for the automatic straightening of a basic turnout rail.