Guided Wave Imaging Based on Fully Connected Neural Network for Quantitative Corrosion Assessment

Abstract Corrosion can affect the reliability of materials, which has attracted the attention of the industry. Corrosion detection and quantitative analysis are particularly important for scientific management and decision-making. In this paper, the imaging method based ultrasonic guided wave (UGW) detection technology and fully connected neural network (FCNN) is proposed to realize real-time imaging of corrosion damages. The imaging method contains offline training and online testing. Offline training aims to establish the relationship between detection signals and velocity maps and it is accelerated by adaptive moment estimation (Adam) algorithm. In the process of online testing, the trained model can be called directly to realize real-time imaging, that is, the detection signals are fed into the model and the network will predict the velocity maps. Finally, the velocity maps are converted to thickness maps according to the dispersion curves. Numerical experimental results show that the mean square errors (mses) are respectively 9.08 × 10−4, 2.47 × 10−3 and 2.59 × 10−3 in training, validation and testing. Compared with irregular corrosion damages, the imaging method has better imaging quality for circular corrosion damages.