Search

This paper presents a new, to the best of our knowledge, methodology for the thermal compensation of background heating in thermograms of composites. The technique analyzes the spatial data of the thermal images obtained from a pulsed thermography inspection and automatically calculates the optimal parameters of a predefined objective function. These parameters are obtained by curve fitting using the least squares method and model the temperature distribution of the image background using the proposed objective function. To verify the methodology, we use real and synthetic images of a sample of carbon-fiber-reinforced plastic (CFRP) with defects, with diameter/depth ratios that range between 15.0 and 75.0 and between 1.7 and 90.0, respectively. The performance of the method is tested using a local and a global definition of the signal-to-noise ratio (SNR) and is statistically validated by analysis of variance. The average performance values obtained were 55.0dB and 7.0dB on synthetic images and real images, respectively. The proposed method provides superior and statistically significant differences compared to techniques reported in the literature for contrast enhancement [e.g., differential absolute contrast (DAC) and background thermal compensation by filtering (BTCF)]. Unlike contrast normalization (CN), the proposed technique stands out since it does not need to predefine variables, select reference regions, have prior knowledge of the partial (or complete) state of the material, or analyze totally (or partially) the temporal evolution of the temperature or any characteristic derived from it.

This paper presents a new methodology for the automatic detection of defective regions of interest (d-ROI) in thermal images of composite materials. The images are acquired with pulsed thermography, and local histograms of oriented gradients are obtained by thermogram processing. This information is analyzed using a simple strategy to differentiate the material background from the defective areas. The procedure is independent of image contrast or enhancement; it does not require analysis of a complete sequence of images, nor does it involve heat transfer models or the extraction of nonuniform heating information. The methodology is tested with synthetic images of a carbon fiber-reinforced plastic sample, containing diameter/depth ratio defects with different values (between 150 and 0.56). The performance of the d-ROI detection method is validated using the area under the ROC curve (AUC) measure, generally obtaining a maximum average value of 0.949 with variations between 0.891 and 0.993 for all the defective depth and size conditions studied. In addition, this method is highly robust when detecting defects in 48.84% of the total number of images, as determined by the sequences analyzed with AUC values higher than 0.95. Outside the high detectability index range, the AUC performance increases abruptly and decays gradually. Recent literature proposes automatic detection of defects in thermograms yielding similar performances to those obtained with the proposed method; however, they require preprocessing of all the thermograms to improve image contrast and visibility and to attenuate the adverse effect of nonuniform heating, which affects the implementation complexity and the computational cost.