SVR-based heterogeneous material structure uniformity multi-parameter ultrasonic characterization method

Abstract

The invention discloses an SVR-based heterogeneous material structure uniformity multi-parameter ultrasonic characterization method, and belongs to the technical field of ultrasonic nondestructive testing and evaluation. According to the method, two parameters of a slope absolute value k and a uniformity length L are extracted by adopting an area fraction multi-scale analysis technology to quantitatively describe composition phase uniformity, wavelet analysis is performed on an ultrasonic detection signal of a heterogeneous material to decouple multi-scale attenuation coefficient characteristics, and a multi-output support vector regression model for quantitatively representing composition phase distribution uniformity is constructed. A genetic algorithm is used for optimizing model hyper-parameters, and ultrasonic collaborative characterization of composition phase distribution uniformity multiple parameters is achieved. The method can provide an effective way for characterizing the distribution uniformity of the microstructure of the heterogeneous material, and has good popularization and application values.

Description

technical field [0001] The invention relates to an SVR-based multi-parameter ultrasonic characterization method for tissue uniformity of heterogeneous materials, belonging to the technical field of ultrasonic nondestructive testing and evaluation of materials. Background technique [0002] Heterogeneous materials are widely used in aerospace, biology, energy and other fields, relying on the synergistic effect of different components in the material to meet their comprehensive high-performance requirements such as heat insulation, wear resistance, wave absorption, and sealing. The uniformity of the microstructure of heterogeneous materials has a very important impact on the improvement of their macroscopic properties and quality control. At present, the quantitative characterization of the uniformity of heterogeneous materials is mostly based on destructive methods, mainly including metallography, scanning electron microscopy, electron probe and X-ray diffraction analysis. T...

AI Technical Summary

Problems solved by technology

[0004] In view of the above problems, the present invention utilizes the excellent performance of the SVR machine learning algorithm in solving problems such as complex nonlinear relationships among multiple parameters or unclear physical mechanisms, and comprehensively considers the sample size, effective mining of ultrasonic information, and the number of target parameters to be measured. Based on factors such as potential correlation and other factors, a genetic algorithm based on ultrasonic time-frequency domain attenuation coefficients is proposed to optimize the multi-output support vector regression (Genetic Algorithm-Multioutput Support Vector Regression, GA-MSVR) model. Compared with single-output support vector regression, this The model can make good use of the correlation between the distribution uniformity parameters of each component phase in the material, and improve the accuracy and robustness of the characterization results while realizing the multi-parameter collaborative characterization of the distribution uniformity

[0005] The present invention provides a multi-parameter ultrasonic characterization method based on SVR for the uniformity of heterogeneous materials, aiming at the strong randomness and differences in the morphology, size and distribution of each constituent phase of heterogeneous materials, which lead to ultrasonic scattering. The mechanism is complex, and the traditional ultrasonic detection method is difficult to simultaneously characterize the distribution uniformity of each component phase of the material. The present invention establishes an ultrasonic quantitative characterization method based on a genetic algorithm optimized multi-output support vector regression model for the distribution uniformity of the component phases.

Images