Machine learning detection of grouting defects in bridge sleeves using impact-echo

Traditional methods for detecting grouting quality inside sleeves using the impact-echo (IE) technique rely heavily on inspectors’ experience, resulting in low efficiency and high subjectivity. To address this challenge, this study proposes a bridge pier grouting defect detection method combining IE signals with machine learning model. First, a hybrid ensemble learning model is developed, taking wavelet packet energy vectors derived from IE signals as input features. Bayesian optimisation is employed to tune the model’s hyper-parameters, enabling accurate identification of grouting defects inside the sleeve. Second, SHAP analysis is applied to interpret the model’s predictions, quantify the contribution of each input feature, and confirm that only a subset of high-importance features is sufficient for high-accuracy prediction. The effectiveness of the proposed method has been validated through field tests on prefabricated bridge piers, demonstrating its potential for engineering applications.