Rail base flaw detection and quantification based on the modal curvature method and the back propagation neural network

Abstract

Due to the development of railways towards high speed and heavy load, a high-precision all-around system for rail flaw detection at the earliest stage is in huge demand. In the published literature, flaws in the rail head and web have already become the primary objective of much research. Fruitful conclusions have been satisfactorily drawn to guide different engineering practices. Unfortunately, flaws developed at the rail base are often overlooked. According to field investigations, the ignorance of this element might sometimes be the main suspect to explain certain derailments and severe accidents. Hence, an efficient technology for detecting rail base flaws is required. This paper investigates the application of the modal curvature method and a neural network. Firstly, the detection and quantification algorithms of rail base flaws are presented. Secondly, a series of numerical studies is introduced on a free rail and a fastened rail, respectively. Modal test results are utilized for verifying the numerical models. Finally, by using the trained neural network, the base flaws are detected and quantified. The proposed algorithm provides a new research idea for the area of rail detection.