Nonlinear Lamb wave-based characterisation of precipitated phase area fraction in creep Ti60 alloy using CNN-BiLSTM-Attention model

Ti60 alloys are widely used in aerospace applications, where high-temperature creep is a major failure mechanism. Under high temperature conditions, Ti60 alloys precipitate silicides, leading to a decrease in the strength, ductility and creep resistance of the material. Monitoring the area fraction of the precipitated phase is essential for evaluating microstructural degradation and ensuring long-term structural integrity. Traditional techniques such as metallography and transmission electron microscopy (TEM) are accurate but destructive and unsuitable for real-time monitoring. This study proposes a deep learning-based method to non-destructively predict the area fraction of precipitated phases in Ti60 alloy after high-temperature creep. The model integrates a multi-scale Convolutional Neural Network (CNN), a Bi-directional Long Short-Term Memory network (BiLSTM) and an attention mechanism. The model uses time-frequency images of nonlinear ultrasonic signals as input, exploiting their sensitivity to microstructural evolution. Experimental results show that the model achieves a prediction accuracy of 98% and a root mean square error (RMSE) of 0.29. The proposed approach offers a rapid, accurate, and non-destructive solution for assessing precipitation-related microstructural changes in Ti60 alloy resulting from high-temperature creep damage.