Ultrasonic Quantitative Characterization Of Seal Coating Component Phase Distribution Uniformity Based On Machine Learning

Seal coating is a sacrificial protective coating that wears itself to ensure the airtightness of airflow channel between the stator and rotor of an aeroengine.Cause seal coating is heterogeneous,the uniform distribution of non-metallic lubricating phase and pores on metal substrate is an important prerequisite for evaluating,regulating and improving the comprehensive performance of the coating.Therefore,the accurate nondestructive quantitative characterization of coating microstructure distribution uniformity is an urgent problem to be solved.Ultrasonic testing has obvious advantages in characterizing the microstructure of materials,but there are still some difficulties in the uniformity characterization of seal coating:(1)the morphology,content,size and distribution of each phase of seal coating are highly random and have significant differences,so it is difficult to accurately describe the distribution uniformity with a single evaluation parameter;(2)the microstructure of seal coating lead to complex ultrasonic scattering mechanism,and the nonlinear relationship between ultrasonic signal characteristics and distribution uniformity parameters of phases in coating,which is difficult to decouple;(3)if multiple parameters are needed to evaluate the distribution uniformity,the existing ultrasonic inversion methods are difficult to achieve collaborative characterization of those parameters.In order to solve the above problems,this paper takes Al Si-PHB seal coating as the research object,and uses the improved Multi-Scale Analysis of Area Fractions(MSAAF)technology to quantitatively describe the distribution uniformity of PHB phase and pores in seal coating.Based on ultrasonic attenuation coefficients in time-frequency domain,a Genetic Algorithm-Multioutput Support Vector Regression(GA-MSVR)model is established to quantitatively characterize the component phase distribution uniformity of seal coating.The main research contents are as follows:(1)Based on the microscopic observation and statistical results of Al Si-PHB seal coatings,25 random multiphase medium models of Al Si-PHB seal coating with 43%PHB and 5%pore were established,which microstructure distribution uniformity is regulated by autocorrelation length.The least square method is used to invert the MSAAF curves,and uniformity scale rate k and uniformity length L_H are used to quantitatively describe the component phase distribution uniformity of seal coating.The results show that as the autocorrelation length increases,the uniformity parameter k decreases and L_H increases of both PHB and pore,which illustrates that the distribution uniformity of the coating component phases decrease;(2)For the established random multiphase medium models of seal coating with different distribution uniformity,the time domain,frequency domain and time-frequency domain attenuation coefficients of the simulation signals are extracted,and the relationships between 3kinds of attenuation coefficients and microstructure distribution uniformity are explored.It is found that under the complex ultrasonic scattering mechanism,the multi-scale time-frequency domain attenuation coefficients can clearly distinguish the seal coatings with different microstructure distribution uniformity;(3)Using the time-frequency domain multi-scale attenuation coefficients of ultrasonic signals from simulation models as input characteristic parameters,a GA-MSVR double-parameter prediction model for distribution uniformity is established based on hypersphere insensitive loss function.Then two parameters k and L_H of PHB and pore can be predicted,respectively.Results show that the determination coefficients between predicted and statistical values of k and L_H are all greater than 0.95,and the relative root mean square errors are all less than 20%.Compared with single-output SVR and BP neural network,GA-MSVR model performs well in double-parameter collaborative prediction,which can quantitatively characterize the distribution uniformity of seal coating component phases accurately.