Indirect identification of ballast bed lateral resistance and automatic adjustment of operation parameters during dynamic stabilization based on Convolutional Neural Networks

In order to make dynamic stabilization operations more automated and intelligent, the primary task is to develop methods for in-track identification and evaluation of track stability during stabilization operations. Furthermore, automatic adjustment methods for operation parameters are needed. This study first constructed a Multi-Body Dynamics (MBD) model of the stabilizer-ballast bed system. The effectiveness of the simulation model was verified using indoor test results. The dynamic responses under various initial ballast conditions and operating parameters were calculated using this model, forming a database of correlated samples between dynamic stabilization parameters, track mechanical parameters, and track vibration responses. Next, a surrogate model based on convolutional neural networks (CNN) was proposed. This model identifies the lateral resistance of the ballast bed using the short-term vibration responses of sleepers and dynamic stabilization parameters, achieving an average error of only 5.27%. Finally, to meet the application demand for automatically adjusting operation parameters to improve stabilization efficiency, a targeted surrogate model was trained. A case was provided to demonstrate the reliability of this method. Through this case study, the error between the lateral resistance obtained by adjusting the operating parameters using the optimization algorithm and the target adjustment value is only 9.25%.