Advancing low-carbon additive manufacturing: an integrated deep learning approach for optimal resource and capacity...

As Additive Manufacturing (AM) shifts towards Make-to-Order (MTO) models, synchronizing raw material inventory with machine capacity becomes critical for handling stochastic demand. However, existing literature often treats inventory control and capacity planning as decoupled problems and frequently overlooks the carbon emissions associated with machine rental logistics. To address these gaps, this study proposes a data-driven integrated decision-making framework for the joint optimization of raw material replenishment and machine leasing. Specifically, we develop an end-to-end hybrid CNN-LSTM deep learning model, where Convolutional Neural Networks (CNN) extract local demand fluctuation patterns and Long Short-Term Memory (LSTM) networks capture long-term temporal dependencies. The model explicitly incorporates carbon emission costs from both material transport and equipment logistics into the objective function.Numerical experiments based on a 550-day simulation using available historical product-demand data demonstrate the superiority of the proposed approach under an expanded benchmark set. The results indicate that the CNN-LSTM strategy achieves a total cost gap of 12.56% relative to the theoretical optimal solution and still outperforms all feasible benchmark policies, including a [Formula: see text]Forecast policy with a 17.20% gap, Base Stock policy with a 29.76% gap, and Constant Order policy with a 43.75% gap. In addition, dedicated ablation experiments show that both LSTM-only and CNN-only variants are substantially inferior to the hybrid architecture. Furthermore, while incurring a moderate increase in carbon emissions of 7.76% over optimal, the proposed model achieves a 100% on-time delivery rate, offering a robust trade-off between economic efficiency, service levels, and environmental sustainability.