Fig. 6: Convolutional encoder-decoder (CED) model generalization.

The cross-trained model is rigorously validated using a second laboratory experiment (p4581) as an independent data set. The input signal is the acoustic emission (AE in red, left axis) and the target signal is the friction derived from the shear stress (σs in blue, right axis) at progressively increasing applied normal loads (3–8 MPa), shown in sequence and delineated by different shading.