In application of the acoustic emission technique (AET) for real-time detection of rail defect, it is essential to unerringly identify the occurrence of the defect via on-line analysis of acoustic emission (AE) signals acquired under working conditions, being exempt from either false-positive or false-negative alarm. Targeting AET-based rail defect detection, this study proposes a pattern recognition method which is formulated using damage-sensitive features extracted from monitoring data. By transferring the acquired AE signal into the frequency domain, moving frequency bands (MFBs), over which the AE burst generated by crack initiation or crack growth in passage of a heavy train is well perceived while the train-induced vibration responses are largely isolated, are first defined. The features extracted over the MFBs are used to construct pattern surfaces characterizing healthy and damaged states of a rail, respectively. A pattern classifier in terms of minimum error rate classification is formulated to define the threshold for discrimination. The proposed method is verified by using the monitoring data acquired by an on-site PZT-based rail switch detection system. The results show that the proposed method can successfully identify the damage state of the monitored rail with the use of AE signals acquired under working conditions of the rail.
展开▼
