1：Experimental Design and Method Selection:

The study used ten different sets of input data for training and verification of the neural network, including raw temperature data, polynomial fitting, principle component analysis, Fourier transform, and others. The neural network was trained to determine defect depth in infrared thermographic nondestructive testing.

2：Sample Selection and Data Sources:

A home-made CFRP sample with five Teflon inserts at different depths was used. The sample was subjected to a classical one-sided TNDT procedure using a halogen lamp, and temperature distributions were recorded as IR image sequences.

3：List of Experimental Equipment and Materials:

A 2.5 kW C-CheckIR halogen lamp from Automation Technology, Germany, and a FLIR GF-309 IR imager operating in the spectral range from 3.8 to 4.05 μm were used.

4：5 kW C-CheckIR halogen lamp from Automation Technology, Germany, and a FLIR GF-309 IR imager operating in the spectral range from 8 to 05 μm were used. Experimental Procedures and Operational Workflow:

4. Experimental Procedures and Operational Workflow: The sample was heated for 10 s, and IR images were captured for 42 s with an acquisition frequency of 15 Hz. The raw IR image sequence was smoothed in both time and space, and various processing algorithms were applied to the data.

5：Data Analysis Methods:

The efficiency of the neural network was evaluated by comparing the relative errors in the evaluation of defect depth using different types of input data.