Energy Consumption Forecasting in 3D Printing Through Enhanced Hybrid Deep Learning Techniques

Resumo

Accurate prediction of energy consumption in fused deposition modeling (FDM) 3D printing is essential for sustainable and cost-efficient additive manufacturing, yet remains challenging due to nonlinear process behavior and time-varying power dynamics. This study presents an enhanced hybrid deep learning framework for forecasting total energy usage in FDM processes by jointly modeling high-resolution power time series and static process parameters. Electrical power data sampled at 1 Hz from 1,260 print jobs were collected using smart power meters and combined with key printing parameters. The proposed architecture integrates convolutional neural networks for temporal feature extraction, an attention-enhanced long short-term memory network to capture long-range dependencies, static feature fusion, and a gradient boosting decision tree regressor. Model performance was evaluated using MAE, RMSE, MAPE, and the coefficient of determination. The hybrid model achieved superior accuracy, with an MAE of 4.12 Wh, RMSE of 5.87 Wh, MAPE of 6.1%, and R² of 0.964, outperforming conventional machine learning and standalone deep learning baselines. Phase-wise and cross-material/printer analyses further demonstrated strong robustness and generalization capability. The results confirm that advanced hybrid deep learning approaches enable reliable and precise energy consumption forecasting in FDM 3D printing, supporting energy-aware process planning and sustainable additive manufacturing practices.