Defects such as debonding, bulges, pores, pits, delaminations, and inclusions in composites commonly occur during manufacturing and service. These defects not only reduce strength and stiffness but also result in structural failures. Reliable non-destructive testing methods are required for evaluating the quality of composite materials. Lock-in thermography (LIT) is a full-field, non-contact, and non-destructive testing method based on image visualization, providing an efficient approach to assessing defect quality. However, the depth resolution of LIT subsurface defects is limited by the excitation frequency. A single excitation frequency can only detect defects within a specific depth range. Thus, if the range of defect depths within the specimen is extensive, inspectors are susceptible to leakage and misdetection of defects when this technique is employed. To overcome the limitations of traditional LIT, we propose a multi-frequency fused method. This method leverages optimal excitation frequency selection, phase extraction, phase enhancement, and phase image fusion to enhance the depth resolution of defects in composite materials. The defect information at different depths within the sample can be integrated into a single fused phase image by employing the proposed algorithm. Meanwhile, this method facilitates clear delineation of defect edges and accurate measurement of defect sizes. Our approach and findings are expected to make significant contributions to both qualitative and quantitative measurements in the non-destructive testing of composite structures.

We put forward a muti-frequency fused LIT method to enhance defect visibility and improve the depth resolution of subsurface defects. This approach consists of four steps: optimal excitation frequency selection, phase extraction, phase enhancement, and image fusion. The optimal thermal wave excitation frequencies and the number of excitation frequencies are initially determined according to the theoretical solution of thermal conduction. The selection of excitation frequencies considers both detection efficiency and quality. Subsequently, the phase images at different frequencies are derived using a correlation algorithm and phase enhancement technique. The phase variable which better reflects the defect information inside the specimen is obtained by transforming the temperature information of the surface during the heating period. The best detection results of defects at different depths within the specimen should be reflected in the phase image corresponding to a specific excitation frequency. Finally, the best detection results of all the defects are integrated into a fused image by adopting a principal component analysis algorithm.

To assess the effectiveness of the proposed method, we conduct an experiment to detect defects of various depths and sizes within glass fiber reinforced polymer (GFRP) laminates. A homemade infrared non-destructive testing system is employed for the experiment. The effectiveness of this method is validated by both qualitative and quantitative analyses, with additional discussion on the influence of experimental parameters. The raw thermal image is shown in Fig. 9(a). Only six defects, which range in depth from 1 to 2 mm and in diameter from 10 to 20 mm and are located in the upper-right corner of the GFRP specimen, can be identified in the raw thermal image due to non-uniform heating. Fig. 7 illustrates the phase images at different excitation frequencies without enhancement processing. Despite significant improvement in non-uniform heating, the contrast of defects remains low, and their edges are blurred due to simple linear stretching. The enhanced phase images at different excitation frequencies are shown in Fig. 8. Enhanced phase images reveal a greater number of defects, although they are distributed across different excitation frequencies due to variations in depth and size. For instance, defects with a depth of 4 mm can only be detected in Figs. 8(c) and (d), while those with a depth of 1-2 mm exhibit higher contrast in Figs. 8(a) and (b). This confirms that the optimal frequency for defect detection correlates with the depth of the defects. Fig. 9(b) shows the fused image. Fifteen defects are detectable, except for one with a diameter of 5 mm and a depth of 4 mm in the lower-left corner, which results in a detection rate of 94%. Additionally, two thermal excitation methods of long pulse thermography and digital frequency modulated thermal wave imaging are also compared. Figure 10 and Table 1 highlight the superiority of the proposed method from qualitative and quantitative perspectives respectively. Figure 11 and Table 2 compare four different image fusion methods, with the principal component analysis method exhibiting the best performance. To balance computational efficiency and detection effectiveness, Figs. 12 and 13 discuss the effect of different thresholds on selecting the optimal excitation frequency. Additionally, the phase difference threshold for this algorithm is determined to be 80% of the peak value.

We introduce a multi-frequency fusion detection method, which involves optimal excitation frequency selection, phase extraction, phase enhancement, and image fusion to improve the depth resolution of subsurface defects and enhance defect contrast. Phase images at different excitation frequencies are extracted by multiple LIT detection and integrated into a fused image. The fused result exhibits greater defect contrast and clearer defect edges than phase images obtained at a single excitation frequency. Additionally, it encompasses information about defects of varying depths within the specimen, thus minimizing misdetection and defect leakage. Experimental results demonstrate the superior detection performance of the proposed method compared to LPT and DFMTWI. Defects with a depth of 4 mm are observable in a sample with a thickness of 5 mm. Furthermore, the influence of critical parameters in the proposed method, such as threshold values, is discussed. The performance of four data fusion algorithms is also evaluated by employing two quantitative image fusion evaluation metrics. The findings suggest that the principal component analysis method is more suitable for the multi-frequency fusion detection strategy. Finally, we provide practical guidance for non-destructive inspection of composite structures.