ISBN 978-3-8439-2653-9

978-3-8439-2653-9, Reihe Elektrotechnik

Soufiene Krimi
Non-Destructive Terahertz Sensor for In-line Contactless Thickness Measurement and Quality Control of Multi-Layered Structures

145 Seiten, Dissertation Technische Universität Kaiserslautern (2016), Softcover, A5

Zusammenfassung / Abstract

Terahertz (THz) time-domain spectroscopy (TDS) is attracting more interest in industrial applications because, besides others, THz technology provides the possibility for contactless and non-destructive testing. In particular, the investigation of multi-layer systems becomes feasible as most automotive paints, ceramics, paper, glass, and plastic are transparent for THz radiation. Using time-of-flight measurements of ultrashort THz pulses, the thicknesses of layer systems can be determined in principle. However, the automotive paint thicknesses usually lie below 50 µm, which are not directly distinguishable for THz waves due to the constructive and destructive superposition of the individual reflections. In order to resolve the thicknesses of the individual layers, a regression approach has to be applied. The approach developed in this work for thickness measurement consists of calibration, simulation and optimization. The known classical calibration models extract the spectral optical material parameters of the investigated coatings from single layers in a dry state. However, the paint layers in real applications are not always deposited on dry coatings. The underlying coating can be dry, wet or in an intermediate state, depending on the coating process. This may lead to inaccuracies in the obtained results. With the novel self-calibration model presented here, one is able to perform the calibration directly on the multilayer system taking into account all effects that may occur during the coating process. By combining the benefits of the new self-calibration model with the simulation method and the stochastic optimization algorithm, a minimum resolution of 4 µm for multilayered automotive paints has been obtained while the known minimum resolution with conventional approaches in the THz range amounts to 18 µm. Furthermore, the evaluation time by using the principle of parallel computing on graphics processing units (GPUs) has been reduced down to 280 ms when investigating four-layer specimens, which corresponds to an acceleration factor of 210 compared to central processing units (CPUs). Hence, a suitable approach for inline, online and offline industrial quality monitoring is provided.