In recent years

ultrasonic guided waves technology has been widely used in nondestructive pipeline detection. However

the weak and insignificant defect echoes caused by the different types of tiny defects such as cracks

void

and dent deformation makes it difficult to identify and classify different types of miero defects. In order to identify the types of different tiny defects

the sensitivity of Duffing system to weak periodie signals was exploited and a signal feature classification method based on the dynamic index fractal dimension of the Duffing system and the back propagation (BP) neural network was proposed. By extracting the fractal dimension、 wavelet cocfficient and time domain signal parameters of the Duffing oscillator after inputting the defect signal to be tested as the characteristic parameters of the echo signal

inputting the BP neural network to complete the construction of the BP neural network

realizing the learning of the weak ultrasonie guided wave signal

classification. The numerical simulation and experimental verification show that the recognition accuracy is significantly improved by taking the fractal dimension of chaos index of three Duffing oscillators into consideration. The accuracy of numerical simulation is increased from 86.35% to 91.85%、 and the accuracy of experimental verification is increased from 83.16% to 86.06%. The numerical simulation and experiment verify that the combination of fractal dimension and BP neural network can effectively improve the identification of pipeline features and defects. The innovative use of fractal as the feature input of BP neural network effectivel y improves the accuracy of classification

facilitating identification and accurate classification

particularly in cases of insufficient experimental data or difficult detection scenarios invol ving small defects in the pipeline. The novel classification method that has been proposed has important significance for the pipeline defects classification and accidents prevention.