Defect detection of composite laminated plates based on ultrasonic lamb waves and deep learning algorithms

To address the defect detection problem of composite laminate, this study proposes a defect defection detection method based on ultrasonic Lamb waves and deep learning algorithms. Taking an anisotropic healthy composite laminate as the base model, simulation models with different defects were generated through batch modifications via code, and the displacement-time data of ultrasonic guided wave propagation in these structures were obtained through simulation analysis. The data were resampled to construct a training dataset with consistent compression lengths while preserving the waveform features of different parameters. Based on this dataset, an active learning strategy was employed to iteratively train an artificial neural network, thereby establishing a surrogate model capable of predicting the corresponding defect waveform from the input defect parameters. By combining this surrogate model with approximate Bayesian inference and subset simulation algorithms, the posterior distribution of defect parameters was inverted from the real waveform signals, and a set of possible defect parameters was obtained through posterior distribution sampling. Analysis of multiple groups of defect inversion data and comparative verification with existing literature methods demonstrate that the proposed method can efficiently and accurately predict the defect parameters of composite laminate, and exhibit significant advantages over other methods in terms of the portability of measurement operations and detection accuracy.