| Due to complex non-linearity nature of inverse thermal problem, TNDE was limited to be a qualitative inspection method for many years. Recently, neural networks have been applied to TNDE to extract quantitative information from recorded infrared images. Neural networks (NN) can handle complex non-lineal problems with access to partially available or noisy data. The quantitative TNDE research works based on NN, which have been carried out by now, are applied to homogenous material such as aluminum or plastic and most of them use experimental data to train suggested network architecture. In this thesis, Quantitative inspection of composite materials such as CFRP is treated by applying NN and neuro-fuzzy approaches. The proposed defect depth estimators and defect detector are trained with simulated data extracted from our numerical heat conduction modeling applied to infrared thermography (IT). Both NN and neuro-fuzzy approaches to quantitative TNDE are tested using simulated and experimental data. |
