Defect features enhancement in long pulse thermography by non-uniformity removal and modal decomposition

Long pulse thermography is a widely used nondestructive testing technique with advantages of full-field measurement and low cost. However, some defect features are often obscured by the non-uniform background as well as noise during the thermal image acquisition process, especially those defects with a small diameter-to-depth ratio (DDR). To address this, a signal-to-noise ratio improvement method is proposed. First, the pristine area is screened out based on the PauTa criterion. Then, real-time dynamic fitting of the background is performed using a simple polynomial model to remove non-uniform background. Finally, an adaptive modal decomposition is performed on the thermal image based on the empirical wavelet transform algorithm to extract defect modes. The method is evaluated through defect detection experiments on a carbon fiber reinforced plastic sample with 12 imitated defects of varying DDRs. Experimental results indicate this method can increase the defect contrast ratio from 16.31 to 65.91 for all defects. For a small defect with a DDR of 2.5:2, where defect features are submerged in noise, this method can also enhance its defect contrast ratio from 0.59 to 9, and so it demonstrates that this method has promise in enhancing the features of small defects.