An absolute stress evaluation method based on the lamb waves time-frequency spectrum and residual network for alleviating the effects of inconsistent coupling conditions

Existing ultrasound-based stress evaluation methods mainly rely on the linear relationship between stress and a single time-domain feature, such as time of arrival (TOA). However, the experiments in this study reveal that adhesive layer inconsistencies can lead to variations in these features, particularly TOA, which may be misinterpreted as stress changes, causing significant errors in absolute stress evaluation. Therefore, a new absolute stress evaluation method that combines the Lamb wave time–frequency spectrum with the residual network (LwTFS-ResNet) is developed, which can alleviate the effect of adhesive layer inconsistencies. This model is pre-trained using a theoretical dataset generated based on the acoustoelasticity theory and a time-domain wave packet model. Once pre-trained, this model requires only a small training dataset measured from a single experimental specimen for final training. Mounting the piezoelectric wafers onto a testing specimen where stress is measured often contains variations related to the adhesive layer. The absolute stress evaluation experiments were conducted using both the LwTFS-ResNet method and the traditional TOA-based method. Adhesive layer inconsistencies caused errors in the TOA-based absolute stress measurement results (max error: 31.40 MPa, root mean square error (RMSE): 29.85 MPa), while the proposed method reduced the max error to 7.83 MPa and RMSE to 4.93 MPa, respectively. This demonstrates that the LwTFS-ResNet effectively mitigates the impact of adhesive layer inconsistencies, significantly improving the accuracy and stability of absolute stress measurements.