Physics-Informed Neural Networks for Defect Detection and Thermal Diffusivity Evaluation in Carbon Fiber-Reinforced Polymer Using Pulsed Thermography

Carbon fiber reinforced polymer (CFRP) is widely used in various industrial applications. However, subsurface defects can compromise the performance and integrity of CFRP products. To enhance quality control and safety, non-destructive testing methods such as active infrared thermography are used for defect detection. In this study, we propose a physics-informed neural network (PINN) that combines experimental data with the priori physical knowledge expressed by Fourier’s law of heat diffusion to process thermographic data. With the help of PINN, non-uniform backgrounds are estimated and removed from the original thermograms, highlighting the defect information. Subsequently, principal component thermography is used to reduce dimensionality and extract features from the processed thermograms. Additionally, PINN can estimate unknown physical parameters such as the material’s thermal diffusivity. We demonstrate the feasibility of the proposed method using experimental and simulated case studies based on pulsed thermography.