热传导逆问题的研究是亚表面缺陷定量评估与识别的理论基础。本文基于非傅里叶热传导模型的正问题研究，即将材料亚表面温度的正问题计算结果作为实验测试数据，采用Levenberg-Marquart算法对半无限材料中圆柱绝热缺陷的逆问题进行了研究。分析讨论了初始假设、入射频率、热扩散长度等对反演结果的影响，并将LM 算法与最速下降法、牛顿法等其他算法进行了比较。结果表明：初始假设参数不影响LM 算法的有效性；入射频率、热扩散长度等物性参数仅会影响收敛速度的快慢，不影响反演结果的精确性；LM 算法相较于最速下降法和牛顿法具有更大的收敛范围和更快的收敛速度。

The research on inverse heat conduction is the theoretical basis of quantitative estimation and identification technology of subsurface defects. An inversion algorithm to characterize depth of subsurface defects in materials is presented. A one-dimensional heat transfer model of semi-infinite material containing a subsurface cylinder was developed and the Levenberg–Marquardt (LM) method was applied to the quantitative detection procedure. The analytical solution for the forward study based on non-Fourier law was employed in the proposed method as the experimental data. The effects of such parameters as initial guess, incident frequency, and thermal diffusion length on the inversion results were examined. Moreover, the comparison between the LM algorithm, the steepest descent method, and the Newton’s method were discussed at the end of the paper to certify the superiority of the LM algorithm. The results of this work are summarized as follows. The reliability and applicability of the LM method applied to pulsed thermography are verified. The inversion accuracy of the results is high in different initial assumption. Thermophysical parameters such as incident frequency and thermal diffusion length do not affect the final inversion results, but the rate of convergence may be affected. Compared with the steepest descent method and the Newton’s method, the LM method has a wider convergence range.