Attention-LSTM for intelligent detection of thermally grown oxide layer using terahertz time-domain spectroscopy

In order to solve the challenge of non-destructive and accurate measurement of the thickness of thermally grown oxide (TGO) and high labor intensity of manual operation, an Attention-Long Short-Term Memory (LSTM) model combining Attention mechanism and LSTM is proposed. The model uses terahertz time-domain spectroscopy (THz-TDS) signal as the input data of TGO non-destructive intelligent detection. Firstly, the thermal barrier coating simulation model with TGO is established by using the finite-difference time-domain method and the simulation signal data of THz-TDS are obtained. Then, the Attention-LSTM model is established. Finally, the simulation data is used to verify the effect of the established model. Experiments show that the mean square error of the model is less than 0.004, the mean absolute error is less than 1.430, the R-square is higher than 0.950, and the average running time is 28 seconds. This method effectively captures the complex relationship between TGO thickness and THz-TDS signal, and provides reliable support for thermal barrier coating life assessment and fault prediction.