Fisher linear discrimination analysis (FLDA) was used to extract signal features and restrain lift off fluctuation in order to solve the problem of pulsed eddy current detection for internal state of multi layer conductive structure and the interference of probe lift off. Then the support vector machine (SVM) method was investigated for multi layer structure internal states automated classification. Detection for internal state of bimetallic temperature controller structure was taken as an example to validate the method by experiment. The theoretical and experimental results show that the proposed method FLDA can realize the maximum separation of the useful information and disturbance information by choosing the best projection direction. FLDA helps to restrain lift off effect and obtain higher recognition accuracy compared with time domain feature extract method based on peak value and peak time. In the case of smaller sample size, the proposed method has a high detection rate