Experimental investigation of the compressive damage characteristics of early-age shotcrete using machine learning and acoustic emission techniques

Abstract

Early-age shotcrete used in tunnel primary linings is often subjected to external loading before adequate hydration has occurred, making the characterization of internal damage evolution critical for structural safety assessment. However, the heterogeneity of early-age shotcrete challenge conventional acoustic emission (AE) source localization methods based on constant wave velocity. In this study, shotcrete specimens cured for 1, 2, and 3 days were tested under uniaxial compression with synchronous AE monitoring. Pencil-lead break tests were conducted to construct a position-known AE dataset, and a three-dimensional AE source localization model based on a Residual-Enhanced Multilayer Perceptron was developed. The model captured the nonlinear relationship between multi-sensor AE features and source coordinates, achieving test-set R² values of 0.9679, 0.9293, and 0.9554 in the X, Y, and Z directions, respectively, with corresponding RMSE values of 3.2677, 4.3899, and 5.0259 mm. With increasing curing age from 1 to 3 days, the mean compressive strength increased from 12.00 to 21.03 MPa, accompanied by a reduction in peak strain from 0.0139 to 0.0098 and elastic modulus increased from 1.98 to 3.17 GPa. AE parameter evolution and AF/RA-based crack classification revealed that 1-day specimens exhibited earlier and more diffuse damage activity, while 3-day specimens showed more concentrated AE energy release near and after the peak stress. The dominant cracking mechanism gradually shifted from initial shear-related interfacial sliding toward tensile opening, followed by mixed tensile–shear failure after peak loading. These findings provide a PLB-calibrated data-driven approach for identifying damage-active regions under the current laboratory configuration and for analyzing the failure mechanisms early-age shotcrete.