Internal defect detection and reconstruction framework for laminated glass fibre reinforced polymer composite materials

The detection of internal defects in composite materials with non-destructive techniques (NDT) is crucial for both quality checks during the production phase and in-service health monitoring during maintenance operations in industrial and civil environment. Visual inspection allows only the analysis of surface characteristics of materials. If internal faults occur inside the composite structures, a deeper analysis is required. Ultrasonic testing has been a promising NDT which is based on the detection and the interpretation of the ultrasonic waves reflected by defects. However, ultrasonic data are difficult to interpret since they require the analysis of continuous signals for each point of the material under consideration. Particularly,the non-homogeneous nature of reinforced composite materials induces high dimensionality of analysis space and a very high level of structural noise that greatlycomplicates the interpretation task. Increasing the ultrasound system frequency can result in detection of smaller defects but the depth of penetration of the wave decreases. Therefore, an advanced signal processmg technique is necessary to manage large data sets and to extract suitable features for effective internal defect detection. The objective ofthe research is to design and develop a new cost-effective nondestructive evaluation (NDE) framework to detect and reconstruct the internal defects in high dimensionality environments. The proposed framework consists of four steps: (i) the relationship between the defects and the behaviour of the ultrasound is identified. (ii) Multiresolution signal decomposition technique is then applied to reduce the dimensionality of the data. (iii) The image ofthe defect region is reconstructed by using the attenuation of the reflected ultrasound signal (iv) Entropy-based fuzzy k-nearest neighbour classification method is used to extract the feature of the defects. Delamination was introduced as the internal defects in the experiments. The proposed framework was tested on glass fibre reinforced polymer (GFRP) composites with different thickness and fiber orientations. The research finding showed that the position of damage has been the significant control factor to the attenuation of the ultrasound signal. Experimental results showed that the proposed framework successfully reduce the dimensionality of the analysis space. The proposed wavelet-based minimization algorithm has achieved 79.8% and 30.2% improvement of signal-to-noise ratio for the simulated and experimental noisy data respectively. This framework exhibits high accuracy of internal defect localization in high dimensionality environments. It is found out that the proposed entropy-based k-nearest neighbour classification method has shown promising performance with 94.0 1% accuracy in close-spaced defects detection when minimal k-nearest neighbor is used. Considering all results and the collected information, it can be concluded that the structural noise in ultrasound signals induces low frequency. Therefore, by removmg the low frequency signals, the intemal defect detectability can be improved. Moreover, the classification of an input pattem based on the closest neighbours of the point of interest provides more accurate defect detection in comparison with the classification based on experience data as the defect pattems vary on circumstances in ultrasonic NDE problems