GraphCompNet: A Position-Aware Model for Predicting and Compensating Shape Deviations in 3D Printing

TL;DR

GraphCompNet presents a position-aware framework for predicting and compensating shape deviations in additive manufacturing using point-cloud inputs and a GAN-inspired two-stage training pipeline. A Dynamic Graph Convolutional Neural Network backbone on isometric meshes captures complex geometries and spatially varying thermal effects to enable closed-loop, batch-consistent compensation. Case studies across nesting scenarios and molded-fiber datasets demonstrate 30–66% improvements in deviation and robustness across positions and orientations, illustrating potential for Digital Twin-enabled industrial AM. The work offers a scalable, real-time capable alternative to physics-heavy simulations, reducing geometry-specific calibration while pushing toward scalable, precision manufacturing.

Abstract

Shape deviation modeling and compensation in additive manufacturing are pivotal for achieving high geometric accuracy and enabling industrial-scale production. Critical challenges persist, including generalizability across complex geometries and adaptability to position-dependent variations in batch production. Traditional methods of controlling geometric deviations often rely on complex parameterized models and repetitive metrology, which can be time-consuming yet not applicable for batch production. In this paper, we present a novel, process-agnostic approach to address the challenge of ensuring geometric precision and accuracy in position-dependent AM production. The proposed GraphCompNet presents a novel computational framework integrating graph-based neural networks with a GAN inspired training paradigm. The framework leverages point cloud representations and dynamic graph convolutional neural networks (DGCNNs) to model intricate geometries while incorporating position-specific thermal and mechanical variations. A two-stage adversarial training process iteratively refines compensated designs using a compensator-predictor architecture, enabling real-time feedback and optimization. Experimental validation across various shapes and positions demonstrates the framework's ability to predict deviations in freeform geometries and adapt to position-dependent batch production conditions, significantly improving compensation accuracy (35 to 65 percent) across the entire printing space, addressing position-dependent variabilities within the print chamber. The proposed method advances the development of a Digital Twin for AM, offering scalable, real-time monitoring and compensation capabilities.

Figures (10)

• Figure 1: Configuration of the Molded Fiber dataset within the printing chamber, illustrating part placement, including stacking in the y-orientation and rotation in the x-orientation. The top row displays the deviations of two identical parts placed at different positions within the same print bucket, with a heatmap indicating the scale of deviation. These parts display distinct deviation patterns across their geometries, as highlighted by the varying color map distributions. The bottom row shows the decreasing deviations after compensation using the proposed compensation framework in both parts.
• Figure 2: The configuration of the bar dataset within the printing chamber (a) illustrates the distribution of identical part geometries placed at different positions for batch production, each labeled with a part ID. The selected training and validation parts for the case studies are highlighted with circles, as shown in (b), a cross-section of the build bed layout.
• Figure 3: The schematic diagram illustrates the proposed architecture for shape deviation prediction and compensation, inspired by the generative adversarial network framework. Initially, the DL compensation engine generates a compensation plan (step 1), which is subsequently evaluated by the DL prediction engine through shape deviation prediction (step 2). Finally, the process iterates as needed, incorporating data-driven feedback (step 3).
• Figure 4: The procedural sequence of the DL prediction engine begins with the conversion of the input CAD into an isometric mesh ($\mathcal{C}_i = \{\mathbf{c}_1, \dots, \mathbf{c}_k\}$). Subsequently, graph neural networks are applied to predict shape deviation ($\hat{\mathcal{S}}_i = \{\hat{\mathbf{s}}_1, \dots, \hat{\mathbf{s}}_k\}$). The accuracy of the predicted shape deviation is assessed by comparing it with the printed and scanned shapes ($S_i = \{\mathbf{s}_1, \dots, \mathbf{s}_k\}$), utilizing a deformation loss function defined in Equation \ref{['eq:total_pred_loss']}.
• Figure 5: The workflow of the DL compensation engine. The DL compensation engine converts the input CAD into isometric mesh ($\mathcal{C}_i$) and then applies graph neural networks to find an optimal compensation plan. It passes through a weight-freeze DL prediction engine, then compares the shape deviation results with the input CAD.