Fig. 4: Instantaneous frictional coefficient prediction from the convolutional encoder-decoder (CED) model trained on Finite-discrete element method (FDEM) simulation data. | Nature Communications

Fig. 4: Instantaneous frictional coefficient prediction from the convolutional encoder-decoder (CED) model trained on Finite-discrete element method (FDEM) simulation data.

From: Predicting fault slip via transfer learning

Fig. 4: Instantaneous frictional coefficient prediction from the convolutional encoder-decoder (CED) model trained on Finite-discrete element method (FDEM) simulation data.

The a input and b predicted scalograms are shown, with the color bar indicating the continuous wavelet transform (CWT) real coefficients. The cross-hatched region in b indicates the cone of influence where edge effects are important. The predictions from the CED are made applying sliding windows with 2 s length and step size of 0.2 s. The predicted scalogram is the average of all sliding windows. c The numerical simulation data (black line) and model-predicted friction coefficient μ from the inverse of the scalogram is shown in red with the blue region indicating 1-standard deviation for the predictions in the overlapping windows. The mean absolute percentage error (MAPE) is listed for the numerical simulation and predicted values.

Back to article page