Abstract: The analysis of ultrasonic signals obtained during weld inspection is dealt with，aiming to distinguish different flaws with signal analysis and classification methods. The features of radio-frequency signal and demodulated signal are described，and an evaluation method of distance and identification index is given，which is used to compare the possibility of being classified. The flaw signal classification is presented on the combination of wavelet analysis and neural network. Results obtain demonstrate the effectiveness of RBFN than BPN in learning speed and generalization.

Key words: Distance between-and-within classes, Flaw classification, Radio-frequency signal, Ultrasonic inspection, Design, Dynamic characteristics, Planetary transmission system, GTF aero-engine