Sparse ultrasonic guided wave imaging with compressive sensing and deep learning. (1st October 2022)

• Record Type: Journal Article
• Authors: Wang, Xiaocen; Li, Jian; Wang, Dingpeng; Huang, Xinjing; Liang, Lin; Tang, Zhifeng; Fan, Zheng; Liu, Yang
• Is Part Of: Mechanical systems and signal processing. Volume 178(2022)
• Publication Date: 2022-10-01
• Subjects: Sparse ultrasonic guided wave imaging -- Compressive sensing -- Deep learning; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2022.109346
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: Sparse UGW imaging with CSDL is proposed and validated in corrosion dataset. The image quality with only a quarter of the transducers is comparable to that of the full transducers. The errors of detection signals from are reduced by 3 to 4 orders of magnitude after data reconstruction. Abstract: A sparse UGW imaging algorithm based on compressive sensing and deep learning models (CSDL) is proposed in this paper to solve the problem that the imaging quality is restricted by the number of transducers in service. The sparse detection signals acquired by a small number of transducers are roughly and finely reconstructed by compressive sensing (CS) and deep learning models. Finally, the reconstructed detection signals are input to the imaging algorithm for sparse imaging. After CSDL, the average correlation coefficients in simulation are respectively improved from 0.9218 and 0.8896 to 0.9626 and 0.9441 with 32 and 16 transducers, which are close to the average correlation coefficient 0.9682 when all 64 transducers are in service. Experimental results also verify the feasibility of CSDL and the average correlation coefficients rise from 0.6091 and 0.6530 to 0.9058 and 0.8357 with 32 transducers and 16 transducers. The proposed method is proven to be effective with only a quarter of transducers in sparse imaging with almost no sacrifice in image quality.