Research On Metal Defect Ultrasonic Intelligent Test Based On Multi-feature Data Fusion

Metal materials will gradually appear cracks, corrosion and other defects in the process of use due to the long accumulation of fatigue and stress, to take a certain safety measures can effectively prevent the sudden fracture of the material, reduce the occurrence of accidents. The evaluation of metal defects in the general engineering field is carried out by the engineer through the eyes and experience to observe the ultrasonic echo signal, which has the disadvantages of low efficiency, poor accuracy and large workload. With the continuous development of artificial intelligence and computer data processing ability, the use of computer aided defect identification is becoming more and more important, which can reduce the workload of staff, improve the identification accuracy, and ensure the consistency of the results of the defect evaluation. Ultrasonic testing signal is a kind of nonlinear and non-stationary signal, which contains many variables, with the structural complexity of detection materials, which is suitable for the analysis of stationary signal has many shortcomings. The feature extraction and selection of the defect echo signal is the premise of the defect recognition, which has a direct influence on the accuracy and reliability of the defect identification. By making flat bottom hole, through hole, flat bottomed slot of defect testing block, with the method of ultrasonic nondestructive testing to obtain the defect echo signal, which carries on the inherent time scale decomposition(ITD), the time domain characteristic parameters and the wavelet packet energy as the defect feature vector are extracted. Sensitivity of neural networks to initial weights and thresholds, using the fruit fly algorithm to optimize the parameters, the identification model is obtained. The corresponding values are obtained by training the recognition of the two characteristic parameters, and the comprehensive recognition rate of each defect is obtained by the D-S theory. The results show that: the method of this paper is close to 100%, which has a very good auxiliary function for the metal defect identification in the engineering field.