ULTRASONIC GUIDED WAVES-BASED NONLINEAR AUTOREGRESSIVE DEFECT DETECTION FOR RAILWAY TRACKS USING FIBER BRAGG GRATING SENSING

Abstract

Structural health monitoring (SHM) is crucial to the maintenance and daily operation of civil infrastructures. Railway system, which plays an important role in modern society, relies heavily on robust monitoring systems to give timely warnings of early-stage defects that potentially could result in the consequences of major traffic incidents, such as derailments. Guided wave testing (GWT) methods have been introduced into the rail track monitoring, featuring long-distance monitoring reliability, high sensitivity, and excellent efficiency. In recent yea rs, the deployment of optical fiber-based GWT on railway system has prevailed traditional piezoelectric sensingbased schemes, due to its reliable performance especially under high electromagnetic interference (EMI) environments. In this paper, experimental studies are conducted, where fiber Bragg grating (FBG) sensors are utilized to receive ultrasonic guided waves (UGWs) on railway tracks, induced by a lead zirconate titanate (PZT) sensor, to detect defects. A narrow-band laser is induced to conduct edge filter demodulation of ultrasonic FBGs, with the sampling frequency of 10 MHz. A nonlinear autoregressive neural network with exogenous inputs (NARX) is trained using the acquired UGW signals and is utilized to evaluate rail track condition by extracting damage sensitive features (DSFs) based on the probabilistic density function (PDF) of the prediction error. First, a DSF baseline is obtained using the UGW data acquired from an intact rail track; then, for an unknown rail condition, the signals are processed by the trained NARX model to calculate DSFs; by comparing the calculated DSFs with the baseline, the rail condition can be evaluated. In this research, various UGW excitation frequencies are deployed, and for each frequency band an individual NARX model is trained. The prediction results show that the proposed method is highly sensitive to rail cracks, with an obvious increase in DSF values when an artificial crack is placed, which denotes the promising application of this method into SHM for mass rail transit systems.