چكيده به لاتين
Today, a better and more intelligent understanding of the condition of railway lines is considered one of the leading issues in this industry, which is operation, safety, the comfort of travel, optimal allocation of resources, and so on. The maintenance of the ballast layer in Iran's railway network has a share of 60% in the maintenance costs, making optimizing strategies and resource allocation inevitable. On the other hand, the most critical factor in the safe running of rail vehicles is track geometry, which helps railway tracks' efficiency and better performance. Therefore, in recent years, the investigation of the relationship between track stiffness (taken from stiffness recording cars) and its changes with the geometrical parameters of the line (taken from track recording cars) has attracted the attention of researchers in the railway fields. This thesis investigates the relationship between high hardness and longitudinal level. In the present study, the analysis of this issue is first based on the data obtained from measuring ballast stiffness and unevenness (longitudinal levels) of a reference test line. Then, to complete the studies, the mentioned position is carried out by conducting a case study and collecting the data obtained from measuring the geometric data and changing the location of the line. It has been investigated in a part of the Tehran-Mashhad railway. In these investigations, after pre-processing of primary data, while analyzing the correlation of data of longitudinal levels and ballast stiffness, the error caused by data prediction with the help of data mining algorithms including linear regression, decision tree, and random forest have been compared. The results indicate a strong and significant relationship between ballast stiffness and longitudinal levels. there is a high correlation between the power spectrum density of the data, especially in the range of 1-4 rad/m wave number (wavelength 1.5 and 3 meters). Also, the results of data analysis by machine learning methods, including linear regression, decision tree, and random forest models, indicate that RMSE is 0.10, 0.14, and 0.12, respectively, which means the accuracy of the linear regression model.
