The operating condition of secondary loop of nuclear power plant has the characteristics that are vulnerable to flow accelerated corrosion phenomena. Because of the flow accelerated corrosion, from 1970 to 2012, in the world 1987 number of events were occurred. [1] Nuclear power plant utilities try to estimate the flow accelerated corrosion induced wall thinning by using CHECWORKS code. CHECWORKS code is based on empirical test results of the pipes. Therefore, CHECWORKS code can only estimate the pipe, which has empirical test result. However, in reality, extract the whole test result from the secondary system is almost impossible. Therefore, for the pipes which are not listed on the CHECWORKS code, ultrasonic measurements were conducted during the maintenance period. For the ultrasonic measure, the insulators in the secondary system should be removed therefore, the measure entails huge works. To overcome this issue, Jung Taek Kim et al. [2] focused on the change of pipes' vibration characteristic due to wall thinning effect. By using vibration signal, pipes thinning condition can be diagnosed in online. Jung Taek Kim used Fourier Transform to analyze vibration characteristics. However, pipes' vibration change was too tiny to classify the differences. By using pre-trained wall thinning classifier, we tried to find possible vibration characteristic. To generate vibration mode, generative adversarial network model is used. After the several training sequences, the generator which is the part of the generative adversarial network imitate vibration data. By combining pre-trained diagnosis network and generator, unknown vibration characteristics may be found. In this study, to estimate pipes' thinning condition several machine learning algorithms (Support vector machine, Convolutional neural network, and Long-short term memory network) were reviewed and applied. Each algorithms were trained by using pipes' vibration signal. As a results, LSTM network shows best classification performance. And also, several vibration modes were imitated by using generative adversarial network.