In order to improve the reliability of detecting the debonding defects at the interface of thermal barrier coatings (TBC), a non-linear frequency modulated (NLFM) infrared thermal wave detection method is proposed. A NLFM infrared thermal wave detection system is built, and zirconia TBC specimens are prepared and tested. The effects of defect diameter, excitation power, initial frequency and termination frequency on the defect detection effect are analyzed. Three algorithms such as principal component analysis (PCA) are used to process the image sequence, and the signal-to-noise ratio (SNR) of each sequence processing algorithm is calculated and compared. The results show that the larger the diameter of the defect, the easier it is to be detected, and the appropriate adjustment of the excitation power or the reduction of the initial and termination frequencies is beneficial to the detection of defects. Compared with the other two algorithms, the PCA method is more effective for image sequence processing. It offers a reference for detecting debonding flaws at the TBC interface.