摘要

Recently the use of ultrasonic guided wave for detecting has become attractive in nondestructive testing. The axial length of the pipe discontinuity is not linear with the guided wave signal amplitude. Thus, it is difficult to identify the degree of damage using the reflection coefficient. In order to explore the relationship between the guided wave signal and the discontinuity length, an ultrasonic processing and pattern recognition tool is studied. The characteristic information of the guided wave signal is extracted by wavelet packet decomposition technique. Several temporal characteristics on each scale are combined to a feature set. In order to reduce the computational complexity, the dimension of the feature set is reduced by the kernel sliced inverse regression (KSIR) algorithm. A pattern recognition model, the continuous hidden markov model (CHMM), is introduced to classify and identify different states of the pipe by pre-trained models. Comparing to the log-likelihood probability calculated by all classifiers, the maximum value corresponds to the right state. A guided wave test to a pipe with artificial corrosion on the outer surface was conducted to verify the correctness of the algorithm. Five discontinuity levels were identified correctly, achieving the objective of preliminary quantitative evaluation. The results show that the CHMM model is more robust and can accurately identify the axial discontinuities.