This paper presents the technology of the ultrasonic TOFD (Time-Of-Flight Diffraction) method for testing welded rail joints in tracks. The specifics of images obtained from such tests and algorithms for acquiring information about defects in rail joints are discussed, treating these images as textures. These algorithms utilize a feature vector extraction procedure for each pixel using wavelet decomposition, which is then optimized using Principal Component Analysis (PCA). In the final stage of image classification, the Support Vector Machines (SVM) method is used as a binary defect classifier.In ultrasonic weld testing, the TOFD (Time-Of-Flight Diffraction) method is increasingly used [15]. It utilizes the phenomenon of ultrasonic wave diffraction at the boundary of a defect. As a result, rather unusual defect images are obtained, requiring additional reconstruction and evaluation procedures.
The natural application for this method is welded rail joints. The first attempts to use the TOFD method for such tests were made in the USA [1]. At PKP (Polish State Railways), welds are tested in a traditional manner (echo or tandem method) using rail flaw detectors or a measuring car [16]. However, satisfactory results are not always obtained here, as irregularities in the running surface of the rail head often occur in the immediate vicinity of the joint, which prevents good acoustic contact of the probes. Furthermore, it would also be advisable to thoroughly examine the area of the lateral parts of the rail foot, as cracks developing from these locations propagate to the remaining parts of the rail, as well as to examine the rail web and head from their lateral surface [11]. (…)
The article contains 21,112 characters.
Source: Logistyka Magazine