Rail crack detection using acoustic emission technique by joint optimization noise clustering and time window feature detection

Abstract Recently, acoustic emission (AE) technology has been investigated to detect rail cracks. However, AE signals of cracks are often submerged in heavy noises in practical application, and these serious noise interferences should be eliminated to obtain a reliable detection result. Based on the joint optimization clustering and time window feature, an improved detection method of rail crack signal is proposed by using AE technology in this paper. The joint optimization method based on Long Short-Term Memory (LSTM) encoder-decoder network and k-means clustering is utilized to achieve a better clustering result of noise signals. Then, the distance thresholds of noise clusters are selected to suppress most of the noise signals. After that, the detection method based on crack duration time feature of time window is further proposed to eliminate false detection and improve the accuracy of crack detection. The detection ability of the proposed method is verified by the signals which are acquired from the real noise environment of railway. Meanwhile, the effectiveness of the proposed method is also demonstrated by comparing with the previous study. The results clearly illustrate that the improved method is effective in detecting rail crack signals under serious noise interference.