Defect characterization in infrared non-destructive testing with learning machines

It is well known that thermal contrast-based quantification methods are strongly affected by the non-uniform heating, the sample shape and the chosen sound area. In this work we propose a reference-free thermal contrast by using the thermal quadrupoles theory and evaluate the limits of defect detection in composite samples by using dynamic principal components analysis (DPCA) and k-nearest neighbor algorithm. Additionally, we propose and validate the radial basis functions (RBF) networks and support vector machines (SVM) for the detection and quantification of defect depth in composite material samples affected by non-uniform heating and with complex shapes.     

Defect detection in aircraft composites by using a neural approach in the analysis of thermographic images

Safety in aeronautics could be improved if continuous checks were guaranteed during the in-service inspection of aircraft. However, until now, the maintenance costs of so doing have proved prohibitive. For this reason there is a great interest for the development of low cost non-destructive inspection techniques that can be applied during normal routine tests. The analysis of the internal defects (not detectable by a visual inspection) of the aircraft composite materials is a difficult task unless invasive techniques are applied. In this paper, we have addressed the problem of inspecting composite materials by using automatic analysis of thermographic techniques. The analysis of the time/space variations in a sequence of thermographic images allows the identification of internal defects in composite materials that otherwise could not be detected. A neural network was trained to extract the information that characterises a range of internal defects in different types of composite materials. After the training phase the same neural network was applied to all the points of a sequence of thermographic images. The experimental results demonstrate the ability of the method to recognize regions containing defects but also to identify the contour regions that cannot be associated either with a defective or with a sound region.     

Detection and classification of flaws in concrete structure using bispectra and neural networks

The problem addressed in this paper is the detection and classification of flaws in concrete structure. It is known that higher-order spectra contain information not present in the power spectrum and can suppress Gaussian noise. Thus estimates of higher-order spectra have been shown to be useful in certain signal processing problems. This paper is concerned with the feature extraction from bispectra for concrete flaw detection. Impact-echo experiments are carried out for three different types of flaw in concrete structure. For each monitoring signal, after bispectral estimation, features are selected from the modules of bispectra in the primary region. For automatic interpretation, a multilayer back-propagation neural network is used as a classifier. Both clean data and data with additive white Gaussian noise are used for training and testing. The classification results obtained experimentally demonstrate that this method has good detection rates in varying environments.