Neural Network Approaches for Defect Detection in Composite Materials

The problem of detecting internal defects in composite materials is felt to be unavoidable in many industrial contexts both for quality control of production lines and for maintenance operations during in-service inspections. Among nondestructive techniques, thermographic image analysis has received much attention for the inspection of composite materials. In this chapter, we address the problem of developing neural network approaches for an automatic system of defect detection by the analysis of sequences of thermographic images. Neural networks are very promising because they offer the opportunity to associate input signals to output classes even in the case of nonlinear mapping. In particular, we have considered supervised and unsupervised approaches for training neural network and we have discussed the pros and cons of their applicability in systems that could help safety inspectors in the diagnosis of problems.