A study of defect depth using neural networks in pulsed phase thermography: modelling, noise, experiments

Abstract

Pulsed phase thermography (PPT) was recently introduced, and up to now analysis of this infrared thermographic approach for non-destructive evaluation has been limited to qualitative aspects. The study presented in this paper is the first attempt to extract quantitative information from PPT results. The approach proposed is based on neural networks well known for their ability to handle complex non-linear problems with access to partial noisy data. In the paper, a thermal model is first presented. This model helps in designing the neural network architecture. PPT fundamentals based on pulsed and lock-in thermography concepts are briefly recalled. Also found in the paper are sections on noise with relations to phase and frequency, neural networks, experimental data on both aluminum and plastic materials. The papers concludes with possible directions of work. The proposed method combining PPT with neural network analysis is shown to be encouraging. The sampling frequency with respect to inspected material thermal conductivity is an experimental limitation.