Stress–strain curves are essential for understanding the mechanical behaviour of materials, particularly in additive manufacturing. However, achieving accurate predictions for material extrusion (MEX) printed components often demands extensive experimental data, while traditional machine learning models struggle to generalise across diverse printing parameters. Here, we introduce a novel long short-term memory and temporal convolutional network with transfer learning (LSTM-TCN-TL) framework that enables accurate stress–strain predictions under varying process conditions using minimal data. Our approach employs transfer learning, adapting models trained on stress–strain data for samples printed at specific angles to predict behaviour across different infill densities. By leveraging transfer learning, the number of required specimens per process parameter is reduced from five to one, achieving an 80 % reduction in experimental workload while maintaining high predictive accuracy. Experimental results demonstrate that the LSTM-TCN model achieves high predictive accuracy on the original dataset (R 2 = 0.94, RMSE = 0.07) and maintains robust performance following transfer learning (R 2 = 0.81, RMSE = 0.09), even when dataset size is significantly reduced, and input variables change from angles to densities. The predicted stress–strain curves closely align with experimental results, highlighting the framework’s generalisability and efficiency. This study provides a scalable and data-efficient solution for stress–strain prediction, reducing data requirements while enhancing model applicability across varying MEX process parameters, thereby advancing predictive modelling capabilities in additive manufacturing.

3-D printing, or additive manufacturing (AM), leverages 3-D computer-aided design models and numerical control to produce objects layer-by-layer, playing a key role in Industry 4.0 and Industry 5.0. Despite its potential to revolutionize manufacturing by creating complex structures more efficiently and cost-effectively, 3-D printing still faces quality issues due to a lack of sufficient data, resulting in improper process parameter settings and poor analyzability. This work introduces a data-driven and physics-assisted machine learning (DP-ML) approach for a 3-D-printed BeltClip object, integrating finite element analysis (FEA) and physics-informed machine learning (PIML). The proposed DP-ML framework provides a cost-effective and time-efficient data collection method using Digimat-AM and a warpage classification algorithm. The data collection begins with obtaining the STereoLithography (STL) file of the BeltClip object from Thingiverse and slicing it in Ultimaker© Cura, considering process parameters such as infill amount, toolpath pattern, layer height, print speed, and extrusion temperature. The resulting G-code file is then input into Digimat-AM for further parameter setting and analysis. In Digimat-AM, glass fiber-filled and unfilled material types are set, undergoing the virtual 3-D printing process, followed by a warpage analysis of the printed BeltClip. The collected 3-D printing data is used to build ML models—deep neural network (DNN), decision tree (DT), support vector machine (SVM), logistic regression (LR), and random forest. The DNN contains three architectures—DNN-1, DNN-2, and DNN-3. Based on the metrics of precision, recall, F1-score, and accuracy, DNN-3 outperforms the others and is chosen for the warpage classification algorithm. The presented DP-ML approach is compared with the state-of-the-art methods and shows a promising capability to predicting warpage, optimizing process parameters, and improving the overall quality and efficiency of a 3-D-printed BeltClip.

Owing to the rapid development in additive manufacturing , the potential to fabricate intricate structures has become a reality, emphasizing the importance of designing structures conducive to additive manufacturing processes . A crucial consideration is the ability to design structures requiring no additional support during manufacturing. This paper employs implicit B-spline representations for self-supporting structure design by integrating a topology optimization model with self-supporting constraints derived analytically from the implicit representation. This analytical derivation for detecting overhang regions enables accurate and efficient calculation of constraints, outperforming other B-spline-based methods. Compared to the traditional voxel-based methods, the implicit B-spline representation significantly expedites the optimization process by reducing the number of design variables. Additionally, several acceleration techniques are implemented to enhance the efficiency of our method, allowing simulations of 3D models with millions of finite elements to be completed within one and half an hour, excelling other B-spline-based methods and voxel-based methods. Various numerical experiments validate its excellent performance, confirming the effectiveness and efficiency of the proposed algorithm.