Algorithm 1 Attention-Enhanced Multi-Scale Hybrid Network Training

Require:   1: train_data: Preprocessed training data   2: train_labels: Corresponding RUL labels   3: val_data: Validation data   4: val_labels: Corresponding RUL labels   5: device: Computational device (‘cpu’ or ‘cuda’)   6: fusion_dim: Dimension for feature fusion   7: learning_rate: Initial learning rate   8: epochs: Number of training epochs   9: batch_size: Batch size for training Ensure:  10: trained_model: Trained neural network model  11: procedure TrainNeuralNetwork($train\_data$, $train\_labels$, $val\_data$, $val\_labels$, $device$, $fusion\_dim$, $learning\_rate$, $epochs$, $batch\_size$)  12:     Initialize CNN-LSTM model with attention mechanisms  13:     Define loss function (e.g., MSE) and optimizer (e.g., AdamW)  14:     for epoch = 1 to epochs do  15:         for batch = 1 to len(train_data)/batch_size do  16:            Load batch data and labels  17:            Forward pass: compute model output  18:            Calculate loss  19:            Backward pass: compute gradients  20:            Update model parameters  21:         end for  22:         Validate model on validation set  23:         Compute validation loss and metrics  24:         Update learning rate scheduler if needed  25:     end forreturn trained_model  26: end procedure