Vashchyshyn L. Detection of defect signals in the magnetic flux leakage diagnostics of railway tracks using wavelet transforms and neural networks.

In the dissertation, the actual scientific and practical problem of detection defects on the initial stages of their development is solved. The task of increasing the velocity of processing of the rail inspection information by automating the process of detecting and distinguishing signals from defects in the magnetic flux leakage diagnostics is also solved. A mother wavelet function for continuous wavelet transform (CWT), which inherits the basic features of the form of the signal from transverse cracks in the rail head was created. It facilitates to locate signals both from developed defects and defects in the early stages of their development. An artificial neural network (ANN) for automatic detection of signals from transverse cracks was constructed. The inputs of ANN are wavelet coefficients obtained by CWT using the created wavelet function. The outputs of ANN submit to the operator of the rail detector car signals that potentially could be caused by defects. It will significantly simplify the operator's work since instead of inspecting the entire flaw signal pattern, he will make expert estimates to only selected fragments. Approaches for increasing the accuracy of detecting signals from the transverse cracks of the rails are proposed. Possible ways of extending ANN to identify other types of defects are also proposed.