NDT coil sensor design and raw signal processing for rail crack detection

• Main Author: Tay, Jian Sheng
• Other Authors: Zheng Yuanjin
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/138985
• Institution: Nanyang Technological University

A triple-coil based electromagnetic-induced thermoacoustic system which is a non-destructive method is proposed to detect flaws inside the rail track. The fundamental of this methodology is derived and simulated using the finite element method (FEM). This report provides the overall structure of this system and the design process is discussed in detail using illustrations and experimental results. The resonant inductive coupling technology is applied during the design of the triple coil where the magnetic field intensity excited by the driving coil is increased. Numerical experiments were conducted to extract the lump parameters of the triple coil for designing the matching network for maximum power transferred from the coil to the rail as well as measuring the scattering parameters to optimize the matching network. A simulation between the triple coil transmitter and the rail is conducted to see the current density directions and distributions. The relationship between the energy density distribution and depth of the rail has concluded that thermal energy mainly concentrates on the surface of the exciting region. An equivalent simplified model of the coil and the rail is shown as an inductive power transferred topology where the efficiency of the power transferred from the coil to the rail is derived which is dependent on the resonant frequency as well as the lift-off distance from the coil to the rail. Experiments were conducted and the induced ultrasonic wave propagating inside the rail with holes and without holes is detected and recorded. These experimental results have concluded that this methodology is feasible which can detect cracks in the rail. Finally, the nature of this methodology enables it to implement automation detection easily since it can produce different types of the ultrasonic signal based on the different types of fault detected inside the rail. These experimental signals are processed through signal processing blocks that extract their unique features. These features are collected and consolidate into training and testing data which are used to train a linear classifier model (Supporting Vector Machine). This supervised machine learning for classification is conducted to classify and make a prediction for new testing data into four different classes mainly: no fault in rail, one hole in rail, three vertical holes in rail as well as three horizontal holes in the rail.