To overcome disadvantage of Levenberg-Marquardt(LM) algorithm,which can not keep geometric correlation of defect parameters,a modified method with fast convergence is presented.In a heat transfer model for a two-dimensional test piece with a subsurface rectangular defect,numerical experiments are conducted to verify effectiveness of the modified LM algorithm.Factors of initial guess and infrared temperature measurement error are considered.Conclusions are drown as follows: The modified LM algorithm keeps geometric correlation of defect parameters well and influence of initial guess is negligible;Identification accuracy of different defect parameters differs due to infrared temperature measurement errors.

A two-dimensional heat transfer model is built and solved with finite element method to study temperature distribution on outer surface of a test piece with a self-heating subsurface defect. Based on conjugate gradient method, an algorithm is proposed for identification of subsurface defect with temperature on the outer surface of the teat piece measured with an infrared imager. Effectiveness of the method is validated by numerical experiments. It is concluded that the method identifies serf-heating defects with high precision for test pieces with small thermal conductivity. Effect of temperature measurement error, number of temperature measurement point and initial defect boundary guess on defect identification can be neglected. The greater the maximum temperature difference on outer test piece surface is, the higher precision of identification is.