### 基于BPNN的封严涂层孔隙分布均匀性超声表征

1. 1. 大连理工大学 无损检测研究所, 大连 116024;
2. 大连理工大学 材料科学与工程学院, 大连 116024
• 收稿日期:2021-01-20 修回日期:2021-02-06 发布日期:2021-04-08
• 通讯作者: 马志远 E-mail:zhiyma@dlut.edu.cn
• 基金资助:
国家自然科学基金（52075078，51805072，51675083）；"兴辽英才计划"项目（XLYC1902082）

### Ultrasonic quantitative characterization of pore distribution uniformity of seal coating based on BPNN

DONG Zhenyi1, LIN Li1, LEI Mingkai2, MA Zhiyuan1

1. 1. Nondestructive Testing & Evaluation Laboratory, Dalian University of Technology, Dalian 116024, China;
2. School of Materials Science and Engineering, Dalian University of Technology, Dalian 116024, China
• Received:2021-01-20 Revised:2021-02-06 Published:2021-04-08
• Supported by:
National Natural Science Foundation of China (52075078, 51805072, 51675083); Program for Excellent Talents in Liaoning (XLYC1902082)

Abstract: The propagation behavior of ultrasonic wave is complex in multi-phase heterogeneous abradable seal coating, so it is difficult to extract effective ultrasonic characteristics and describe the relationship between the characteristics and microstructure distribution uniformity. The BP Neural Network (BPNN) combined with wavelet transform is proposed to quantitatively characterize pore distribution uniformity of abradable seal coating through the ultrasonic testing technique. Based on the statistical method and the random medium theory, random multiphase medium models for AlSi-Polyester (AlSi-PHB) seal coating were established, whose pore distributions were adjusted by varying autocorrelation lengths. Two uniformity parameters, absolute slope, and uniformity length were extracted to quantitatively describe pore distribution uniformity using the multi-scale analysis of area fraction. The time domain, frequency domain, and time-frequency domain attenuation coefficients were extracted from simulated ultrasonic signals. Correlations between absolute slope, uniformity length and the attenuation coefficients were explored. The BPNN with attenuation coefficients as the inputs was applied for training and prediction of absolute slope and uniformity length. The results show that when porosity was 5%, as autocorrelation lengths increased from 11 μm to 26 μm, absolute slope decreased from 0.86 to 0.76 and uniformity length increased from 550 μm to 6 785 μm, which indicates that the pore distribution uniformity gets worse. The time-frequency domain attenuation coefficients and the uniformity parameters were the most correlative, and the determination coefficients of the predicted absolute slope and uniformity length through BPNN were 0.97 and 0.98, respectively, which are the most accurate among the prediction results of three kinds of attenuation coefficients. The proposed method provides an effective way for quantitative characterization of distribution uniformity of multi-phase heterogeneous coatings.