Comparing the efficiency of defect depth characterization algorithms in the inspection of CFRP by using one-sided pulsed thermal NDT

•Eight algorithms of defect depth characterization have been comparatively analyzed.•Universal characterization technique is pulse phase thermography.•The technique of TSR is weakly dependent on uneven heating and lateral heat diffusion.•Nonlinear fitting allows characterization of defects in CFRP at depths less than 2 mm.•Neural networks provide the most efficient defect characterization. The efficiency of eight algorithms of defect depth characterization (pulse phase thermography – PPT, thermographic signal reconstruction by analyzing the first and second derivatives– TSR, early observation – EO, apparent thermal inertia – ATI, thermal quadrupoles - TQ, non-linear fitting - NLF and neural networks – NN) has been comparatively analyzed on both theoretical and experimental IR image sequences obtained in the inspection of CFRP composite. Synthetic noise-free image sequences have been calculated by means of the ThermoCalc-3D software, while experimental results have been obtained by applying a one-sided procedure of pulsed thermal NDT to the inspection of artificial defects in CFRP. A relative error in the evaluation of defect depth has been chosen as a figure of merit. It has been demonstrated that a simple and robust processing technique is the use of the Fourier transform resulting in phase-domain data (PPT). The technique of TSR ensures maximal values of signal-to-noise ratio and is less susceptible to uneven heating and lateral heat diffusion. The calculation of ATI has allowed the characterization of defects at depths up to 1.5 mm, but it is sensitive to uneven heating thus requiring to carefully choose a non-defect area. The EO method, as well as the technique of TQ, have revealed inferior results in defect depth identification because of a noisy character of raw signals. Non-linear fitting is a convenient processing technique allowing simultaneous characterization of some test parameters, such as material thermal properties, defect depth and thickness, etc., but this technique is time-consuming and can hardly be applied to full-format images. In the whole defect depth range, minimal characterization errors have been ensured by the use of the NN that is a promising tool for automated identification of hidden defects.