Defect detection in thermography non-destructive testing by means of signal processing and K-Means clustering

This study investigates the application of two Thermography non-destructive testing (TNDT) techniques, pulsed thermography (PT) and step heating thermography (SHT), as well as signal processing techniques for defect detection in materials with varying thermophysical properties. Three thermal signal processing techniques, namely principal component analysis (PCA), thermographic signal reconstruction (TSR), and total harmonic distortion (THD) were employed to enhance defect detectability. In addition, a novel combination of TSR and THD was introduced. The comparative analysis using K-means clustering revealed that PT was more effective in aluminium alloy, detecting 12 (of 20) defects compared to 11 (of 20) in steel due to aluminium’s higher thermal diffusivity. In contrast, SHT proved to be more effective in steel, uncovering 14 defects versus 4 in aluminium, as SHT is better suited for materials with lower thermal diffusivity. Interestingly, THD negatively impacted defect detection with flash excitation, revealing 11 and 8 defects in aluminium alloy and steel, respectively. However, THD with step heating detected 6 defects in aluminium and 15 defects in steel. The first derivative method proved to be effective in both PT and SHT. Notably, the combination of TSR and THD exhibited excellent performance with the step heating method and acceptable performance with PT.