Defects such as debonding, bulges, pores, pits, delaminations, and inclusions in composites are common during manufacture and service. They not only reduce strength and stiffness but also fail structures. Reliable non-destructive testing methods are required to assess the quality of composite materials. Long pulse thermography (LPT) is a full-field, non-contact, and non-destructive testing method based on image visualization that provides an efficient way to assess the defect quality. However, the defect visibility of LPT can be compromised by various factors such as experimental conditions, heat intensity, inherent material properties, and noise. The LPT effectiveness is constrained by fuzzy edges and low-contrast defects. Consequently, enhancing defect visibility via signal processing methods is crucial for inspecting defects in composite materials using LPT. Thus, we propose an infrared image sequence processing method that utilizes Fourier transform, phase integration, and edge-preserving filters to enhance the quality of LPT detection results for composite materials. Meanwhile, a few latent variables that better reflect the defect information inside the specimen are proposed by transforming the temperature information of the surface during the cooling period. These variables can eliminate the influence of uneven heating and improve defect visualization. This method enables clear delineation of defect edges and accurate measurement of defect sizes. Our approach and findings are expected to contribute to qualitative and quantitative measurements in the non-destructive testing of composite structures.

We propose a novel infrared image sequence processing algorithm to enhance the defect visibility of LPT. This approach comprises four steps of background uniformity processing, phase extraction, frequency domain integration, and image quantization. Initially, thermal data is acquired after a square pulse heating period and subsequently pre-processed to eliminate the inhomogeneity of the initial temperature distribution. Subsequently, phase Fourier analysis is conducted to extract the phase information related to defects of varying depths and sizes. Next, the phase difference between defect and sound regions is pixel-wise integrated along frequencies to integrate defect information into a new image. Lastly, the integrated phase image transforms into an 8-bit visual image by applying edge-preserving filters and local adaptive Gamma correction.

To evaluate the effectiveness of the proposed method, we conduct an experiment using a glass fiber reinforced polymer (GFRP) panel and compare it with various thermal signal processing methods. The efficacy of the proposed method is substantiated via qualitative and quantitative analysis, with the influence of acquisition parameters additionally discussed. Figure 7 illustrates the raw infrared images captured in different instances. The defects with deep depths have low contrast and fuzzy edges. The phase images processed by background uniformity and Fourier transform are depicted in Figs. 9(a)-9(c). The visibility of defects in these phase images is improved compared to the raw images. However, the deeper defects are more obvious in the phase images at low frequencies and vice versa. It is challenging to identify all defects at various depths using only phase images at a single frequency. To this end, the frequency domain integration method is utilized to amalgamate the phase information of all defects, and subsequently, the resulting phase integration image is enhanced and quantified. The processed results are presented in Fig. 9(d), where all 20 defects of various depths and sizes are distinguishable. The edges of the defects are visible, which facilitates subsequent image segmentation and edge extraction processing for accurate defect size measurement. Additionally, three traditional thermal signal processing algorithms of absolute thermal contrast, thermographic signal reconstruction, and principal component analysis are also compared. Figures 11 and 12 highlight the superiority of the proposed method from qualitative and quantitative perspectives respectively. Analyzing the variations in temperature difference over time and the signal-to-noise ratio across various sampling frequencies (Fig. 13) allows for determining the optimal acquisition time of 30 seconds and a sampling frequency of 30 Hz, striking a balance between computational efficiency and detection effectiveness.

We employ a homemade infrared non-destructive testing system utilizing LPT for the experiments. A method for processing infrared image sequences based on Fourier transform, phase integration, and edge-preserving filters is developed to mitigate the influence of uneven heating and enhance the contrast of defects. The inspection results of the GFRP panel demonstrate that phase signals can offer more information about defects, and integrating phase information across all frequencies significantly enhances detection performance compared to a fixed-frequency signal phase image. Meanwhile, the accurate defect size measurement in segmented images further validates the reliability of the proposed method. An important advantage of this method is that fewer parameters should be determined, specifically the optimum sampling time and frame rate. Other data dimensionality reduction techniques such as ATC, TSR, or PCA can yield multiple principal component images requiring human visual interpretation. In contrast, the proposed method generates a single optimal detection image, which significantly amplifies the detection automation. Finally, our study provides guidance for practical non-destructive inspection of composite structures.