Layered stiffness detection of ballastless track based on loading force and multiple displacements

Abstract

Grasping the track stiffness status is significant to railway maintenance. However, the research on the data collection and detection method of ballastless track layered stiffness is lacking and challenging. This article proposes a data collection strategy for layered stiffness detection based on loading force and multiple displacements. The dataset, which consists of loading force and multiple displacements collected along the railway line, effectively reflects track layered stiffness, including the overall track stiffness and the slab upper and bottom stiffness. The stiffness detection data is input into the BP model optimized by particle swarm optimization (PSO-BP) to mine the correlation between different sublayer defects, track layered stiffness' fluctuation, and then predict the layered stiffness sequences and locate local anomalies. On this basis, an image dataset of 25 abnormal layered stiffness cases is constructed, caused by different degrees of abnormal fastener stiffness, mortar void, subgrade settlement and their overlap. The Resnnet18 model, pre-trained by transfer learning, is used to identify layered stiffness anomaly cases in image datasets, and the accuracy is 94.63 %.