Deep emulational semi-supervised knowledge probability imaging method for plate structural health monitoring using guided waves

Deep networks can obtain the structural state features and optimize the parameters of the feature layer according to the training labels. The training data including the damage signals are quite helpful for detection model training, but sometimes the industrial damage signals are difficult to obtain, especially in airplane skin and other large structures. In this paper, a deep emulational semi-supervised probability imaging algorithm is proposed to present the damage state in the absence of damage samples. A promising signal generation method for simulated damage was implemented through signal encoding, ReLU activation and reconstruction with disturbance, and its effectiveness was verified in metal plate structures and anisotropic composite plate structures. The experiment results illustrate that the proposed method can detect the damage only using normal state signals, presents a good materials generalization in both aluminium plate and composite plate, and has better performance than other state-of-art methods.

Automated summary

In this paper, a deep emulational semi-supervised probability imaging algorithm is proposed to present the damage state in the absence of damage samples, and its effectiveness is verified through experiments.