Abstract
Vacuum Assisted Resin Transfer Molding (VARTM) is a popular manufacturing method of composite materials, where the porous preform placed in a single sided mold is impregnated with thermoset resin through the use of vacuum. It has been applied in many fields, including manufacturing carbon fiber reinforced polymers (CFRP). In the recent years, CFRP has been utilized due to its high strength and light weight. Nevertheless, the price of CFRP is relatively high. As a result, if the CFRP products contain irremediable defects, the economic loss is inevitably large. Therefore, efficient Non-Destructive Testing (NDT) is desired to ensure the quality of products. Pulsed thermograph (PT) is a popular NDT technique for the convenient deployment and rapid detection. However, as PT data is usually acquired rapidly, the information of flaws is often contaminated by a variety of disturbances. For this reason, a number of data processing methods have been developed to enhance the detectability of PT through eliminating the effects of non-uniform heating and measurement noise. Conventional thermographic image analysis techniques include Pulsed Phase thermography (PPT), Thermal signal Reconstruction (TSR), et al. However, these methods cannot completely remove the influence of the interference in thermal images. Hence, two statistical algorithms are proposed to counter this problem, including Multi-dimensional Ensemble Empirical Mode Decomposition (MEEMD) and Penalized Least squares methods. In this thesis, first, a VARTM experiment system was constructed to manufacture the CFRP specimens, while a PT system was established to acquire thermographic data. Then, PPT, TSR, MEEMD, and AIRPLS were utilized for thermal image processing and compared through Signal-to-Noise Ratio (SNR). The results show that the proposed MEEMD and AIRPLS methods significantly outperform the conventional methods by better eliminating non-uniform backgrounds and noise contained in thermal images.