Efficient Improved SSA for 3D Geometric Information Recovery of Surface Defect in Eddy Current NDT

Abstract

The quality of key equipment is important during its lifelong maintenance. Non-destructive testing (NDT) is a non-invasive technique which ensure the completeness of the equipment during the quality evaluation. In this article, an improved sparrow search algorithm (ISSA) is proposed for efficient 3D surface defect reconstruction in eddy current NDT. The proposed method optimizes the computational efficiency over the traditional SSA, which has the limitations that some individuals conduct redundant or invalid searches. It reduces the search efficiency of the algorithm in the face of high-dimensional search spaces during complex 3D surface defect reconstruction, which directly leads to the prolonged inversion time costs. To address this issue, the ISSA introduces a variable helix factor, adaptive weight, and improved boundary constraints, which aim at strengthening the scroungers’ global and local search capabilities at different stages, enhancing the population’s overall optimization ability, and improving the search effectiveness of boundary individuals. These improvements boost the search efficiency in high-dimensional spaces, enabling the method to achieve more efficient solutions to 3D inverse problems in ECNDT. Furthermore, for high-dimensional input features, the well-known full-factorial sampling (FFS) strategy requires many samples, which leads to high computational burdens in constructing the relationship between the responses and geometric parameters of the surface defects, whereas the Latin hypercube sampling (LHS) strategy effectively overcomes this issue by adopting a stratified sampling mechanism. Both the prediction accuracy and computational efficiency of the proposed method are validated by comparing the predicted defect geometric sizes with the ground truth through simulations and experiments. The results show that the proposed ISSA-based inverse solver provides a reliable solution for surface defect reconstruction in 3D ECNDT.