In additive manufacturing (AM), each change in part geometry requires a new thermal simulation, which makes design evaluation over many candidates costly. Geometry-aware models reduce this cost by predicting thermal fields on unseen parts with a single learned model, but they assume that diverse geometries are available centrally. This assumption is restrictive when each manufacturing site holds only a limited set of proprietary part designs that it cannot share. We propose Fed-PIGANO, a federated framework that trains physics-informed geometry-aware neural operator (PI-GANO) models at individual sites and aggregates their geometry encoders and prediction networks without sharing part geometries. We evaluate the framework on a 2D steady-state thermal benchmark for metal AM, with geometries from five polygonal families distributed across three sites under four scenarios that range from identical to strongly different family compositions. Over seven replications, Fed-PIGANO achieves mean relative L2 errors of 1.83% to 2.04% across all 12 scenario-site cases, compared with up to 20.46% for local-only training, and consistently yields lower PDE residuals. Its prediction accuracy is comparable to that of a centralized model trained on the pooled geometries of all sites. These results demonstrate that federated learning lets manufacturing sites benefit from each other's geometries without sharing them, enabling rapid thermal prediction on new part designs for tasks such as shape optimization, inverse design, and uncertainty quantification.
Figures & tables
Figure 1: Distributed-data challenge in AM. (a) Each of the C sites holds a limited set of nc part geometries. (b) Local-only training keeps data private but covers few geometries. (c) Pooling broadens coverage but may be restricted by data-sharing constraints.
Study
Physics- informed
Geometry- aware
Federated learning
AM thermal focus
PIPN ( Kashefi and Mukerji, 2022 )
✓
✓
×
×
PI-KAN-PointNet ( Kashefi and Mukerji, 2026 )
✓
✓
×
×
PI-GANO ( Zhong and Meidani, 2025 )
✓
✓
×
×
coPINN ( Gnanasambandam et al., 2026 )
✓
×
×
✓
GeoPIPN ( Saimon et al., 2026a )
✓
✓
×
✓
FedAvg-Adam ( Zhang et al., 2025a )
✓
×
✓
×
Table 1: Comparison of related studies and the research gap addressed in this work.
Parameter
Symbol
Value
Parameter
Symbol
Value
Thermal conductivity
k
9.8 W m -1 K -1
Laser power
P
375 W
Ambient temperature
Ta
300 K
Absorptivity
α
0.4
Effective layer thickness
h
2.5 mm
Absorbed power
Qtotal=αP
150 W
Beam radius
rb
2.5 mm
Source center
(xL,yL)
(−1.6rb,0)
Table 2: Material and process parameters of the thermal model.
Figure 2: Fed-PIGANO framework. Each site c∈{1,…,C} trains a local model θc that maps Xb and (x,y) to Tθ , using only its own geometries. In each round, the server aggregates all model components into θr=∑cwcθcr and broadcasts it back. The trained global model predicts thermal fields on unseen geometries.
Figure 3: Geometry generation scheme for the five families. Orange polygons show one example geometry per family.
Scenario
Split
Client 1
Client 2
Client 3
S1 ( δ=0.000 )
Train (50)
(10, 10, 10, 10, 10)
(10, 10, 10, 10, 10)
(10, 10, 10, 10, 10)
Val (25)
( 0 5, 0 5, 0 5, 0 5, 0 5)
( 0 5, 0 5, 0 5, 0 5, 0 5)
( 0 5, 0 5, 0 5, 0 5, 0 5)
Test (50)
(10, 10, 10, 10, 10)
(10, 10, 10, 10, 10)
(10, 10, 10, 10, 10)
S2 ( δ=0.198 )
Train (50)
(22, 16, 0 6, 0 4, 0 2)
( 0 4, 10, 18, 12, 0 6)
( 0 4, 0 4, 0 6, 14, 22)
Val (25)
(11, 0 8, 0 3, 0 2, 0 1)
( 0 2, 0 5, 0 9, 0 6, 0 3)
( 0 2, 0 2, 0 3, 0 7, 11)
Test (50)
(10, 10, 10, 10, 10)
(10, 10, 10, 10, 10)
(10, 10, 10, 10, 10)
Table 3: Per-client family counts (Q, P, H, CP, CH) of the training, validation, and test sets in each scenario. Larger δ (Eq. ( 20 )) indicates stronger heterogeneity across clients.
Table 4: Training settings of the three models. Shared settings span all three columns.
Client 1
Client 2
Client 3
Method
ϵrel↓ (%)
ϵPDE↓ ( ×10−2 )
ϵrel↓ (%)
ϵPDE↓ ( ×10−2 )
ϵrel↓ (%)
ϵPDE↓ ( ×10−2 )
S1 ( δ=0.000 )
Local-only
4.36±0.41
6.55±0.77
4.25±0.53
6.88±0.65
4.34±0.40
6.43±0.65
Fed-PIGANO
1.91±0.17
3.34±0.25
1.83±0.15
3.40±0.35
1.89±0.22
3.45±0.28
Centralized †
1.91±0.10
2.77±0.15
1.82±0.10
2.79±0.19
1.89±0.15
2.87±0.21
S2 ( δ=0.198 )
Table 5: Relative ℓ2 error and PDE-residual RMSE on unseen test geometries for each scenario and client, as mean ± standard deviation over seven replications. Bold marks the lower mean of Local-only and Fed-PIGANO.
Figure 4: Mean test-set ϵrel (top) and ϵPDE (bottom) versus δ for each client (columns). Error bars show ±1 standard deviation over the seven replications.
Figure 5: Mean ϵrel across clients for (a) S1 to (d) S4. Error bars show ±1 standard deviation over the 50 test geometries. Lower error bars are cut at 20% of the bar height where the standard deviation exceeds 80% of the mean.
Figure 6: Mean ϵPDE across clients for (a) S1 to (d) S4. Error bars as in Fig. 5 .
Figure 7: Geometry-wise ϵrel (top row of each subfigure) and ϵPDE (bottom row) of Fed-PIGANO versus Local-only for (a) S1 to (d) S4. Markers below the identity line favor Fed-PIGANO.
Figure 8: Thermal predictions for the unseen CH geometry #45 at Client 1 in S4. The left panel shows the FEM reference. The rows show the predicted temperature, absolute error, and absolute PDE residual of each model. Error and residual color scales are shared across Figs. 8 – 10 .
Figure 9: Thermal predictions for the unseen CH geometry #49 at Client 2 in S4. Layout as in Fig. 8 .
Figure 10: Thermal predictions for the unseen P geometry #16 at Client 3 in S4. Layout as in Fig. 8 .
Accurate thermal modeling is essential in metal additive manufacturing (AM) for understanding the process-structure-property chain. Physics-informed neural networks (PINNs) offer effective surrogate thermal modeling by minimizing physics-based residual losses at collocation points. However, prior works typically rely on manually-crafted, static collocation sampling strategies, which are neither principled nor scalable across process conditions, hindering their generalization capability. In this work, we provide theoretical analysis through empirical risk minimization, showing that process condition-aware adaptive sampling is strictly more favorable than conventional static sampling for generalization. Building on this insight, we propose an adaptive sampling strategy within a two-stage framework: (1) a conditional Flow Matching model that learns approximate high-residual distributions across different process conditions, and (2) a mixed sampling strategy combining this distribution with a domain-informed base distribution to generate adaptive collocation points for refining the PINN predictor. Experiments on metal AM numerical benchmarks demonstrate that our method consistently outperforms state-of-the-art PINN baselines, achieving an average 62.1% reduction in relative L2 error under an identical collocation budget, by capturing process-dependent heat dissipation regions often overlooked in the literature. To the authors' knowledge, this is the first adaptive sampling strategy for PINNs in metal AM, contributing to the enhanced generalization and broader applicability.
Hyeonsu Lee, Jihoon Jeong
Wm Michael Barnes ’64 Department of Industrial and Systems Engineering Texas A&M University College Station, TX 77843, USA
Accurate thermal modeling in metal additive manufacturing (AM) is essential for understanding the process-structure-performance relationship. While prior studies have explored generalization across unseen process conditions, they often require extensive datasets, costly retraining, or pre-training. Generalization across different materials also remains relatively unexplored due to the challenges posed by distinct material-dependent thermal behaviors. This paper introduces a parametric physics-informed neural network (PINN) framework for zero-shot generalization across arbitrary materials without labeled data, retraining, or pre-training. The framework adopts a decoupled parametric PINN architecture that separately encodes material properties and spatiotemporal coordinates, fusing them through conditional modulation to better align with the multiplicative role of material parameters in the governing equation and boundary conditions. Physics-guided output scaling derived from Rosenthal's analytical solution and a hybrid optimization strategy are further incorporated to enhance physical consistency, training stability, and convergence. Experiments on bare plate laser powder bed fusion (LPBF) across diverse metal alloys, including both in-distribution and out-of-distribution cases, demonstrate effective zero-shot generalizability along with superior training efficiency. Specifically, the proposed framework achieved up to a 64.2% reduction in relative L2 error compared to the non-parametric baseline while surpassing its performance within only 4.4% of the baseline training epochs. Ablation studies confirm that the proposed framework's components are broadly applicable to other PINN-based approaches. Overall, the proposed framework provides an efficient and scalable material-agnostic solution for zero-shot thermal modeling, contributing to more flexible and practical deployment in metal AM.
Hyeonsu Lee, Jihoon Jeong
Wm Michael Barnes ’64 Department of Industrial and Systems Engineering Texas A&M University College Station, TX 77843, USA
A comprehensive understanding of heat transport is essential for optimizing various mechanical and engineering applications, including 3D printing. Recent advances in machine learning, combined with physics-based models, have enabled a powerful fusion of numerical methods and data-driven algorithms. This progress is driven by the availability of limited sensor data in various engineering and scientific domains, where the cost of data collection and the inaccessibility of certain measurements are high. To this end, we present PiGRAND, a Physics-informed graph neural diffusion framework. In order to reduce the computational complexity of graph learning, an efficient graph construction procedure was developed. Our approach is inspired by the explicit Euler and implicit Crank-Nicolson methods for modeling continuous heat transport, leveraging sub-learning models to secure the accurate diffusion across graph nodes. To enhance computational performance, our approach is combined with efficient transfer learning. We evaluate PiGRAND on thermal images from 3D printing, demonstrating significant improvements in prediction accuracy and computational performance compared to traditional graph neural diffusion (GRAND) and physics-informed neural networks (PINNs). These enhancements are attributed to the incorporation of physical principles derived from the theoretical study of partial differential equations (PDEs) into the learning model. The PiGRAND code is open-sourced on GitHub: https://github.com/bu32loxa/PiGRAND
Benjamin Uhrich, Tim Häntschel, Erhard Rahm
Center for Scalable Data Analytics and Artificial Intelligence Dresden/Leipzig, Germany. · Leipzig University, Leipzig, Germany.