Automated procedure for detecting and characterizing defects in GFRP composite by using thermal nondestructive testing.

• A neural network can be trained by both experimental and theoretical data. • Experimental input data ensures better defect identification than modeling data. • Automated procedure provides characterization accuracy up to 15% by defect depth. • Experimental TSR profiles provide low scattering of characterization results. The paper describes the concept of an automated defect characterization procedure by using infrared nondestructive testing of glass fiber reinforced composite. The proposed algorithms have allowed determination of defect depth, lateral dimensions and area, as well as coordinates of defect centers. The algorithms are based on the use of the neural network trained on both experimental and theoretical temperature profiles. An acceptable for practice accuracy of defect characterization has been obtained on the experimental data (0–15% by defect depth and 26–139% by defect area). [ABSTRACT FROM AUTHOR]