Defect characterization in pulsed thermography: a statistical method compared with Kohonen and Perceptron neural networks

Pulsed thermography is a popular NDE technique. In this paper, a novel statistical processing method is proposed and compared in term of defect detection and characterization with two common neural network architectures (Perceptron and Kohonen). Interest is on characterization of aluminum corrosion. The statistical method and neural network architectures use temperature, phase and amplitude data with phase and amplitude data coming from the so-called pulsed phase thermography approach. The statistical method reveals interesting performance over tested neural networks, especially in the ’interference technique,’ a combined ’two-step’ approach: detection with phase and characterization with amplitude. Theory is discussed and examples of results are presented.