De-Risking Robotic Welding for In-Space Metal Joining Through Physics-Grounded Digital Twins
Authors: Zhou Daniel Hao, Umer Masood Chaudry, Robert Shaw, Andrew Cheney, Nick Ludford, Mark Sims, Hongbiao Dong
Organizations: DANiLab, University of Leicester, Leicester, United Kingdom · School of Computing and Mathematical Sciences, University of Leicester, Leicester, United Kingdom · TWI Ltd, United Kingdom · School of Metallurgy and Materials, University of Birmingham, Birmingham, United Kingdom
Growing interest in in-space servicing, manufacturing, and assembly has renewed attention on reliable metal joining for modular construction and on-orbit repair. Robotic welding is promising, but reduced-gravity and low-pressure operation remains poorly characterised and flight-like tests are limited. The UK Space Agency-supported ISPARK project is developing a physics-grounded digital twin for structured process de-risking. This paper evaluates a model-based prototype combining a finite-volume thermal weld-pool model, Gaussian-process (GP) and neural-network surrogates, and a simulated six-axis manipulator. The reduced model includes solid-liquid phase change, effective thermocapillary transport, and a local-pressure parameter. Gravity is assessed separately through a hydrostatic-to-capillary pressure ratio and does not enter the thermal energy equation. Across repeated five-fold cross-validation, the GP has lower normalised mean absolute error than the neural network for all eight outputs. In 10 held-out boundary cases, 74 of 80 solver responses fall within nominal 90% GP predictive intervals. A three-output GP query takes 249 microseconds, including scaling and predictive variance. In a simulated UR5 L-shaped-seam study, a proportional controller adjusts the synergic power-source setting as travel speed falls near the corner. Finite-volume re-evaluation finds all 12 closed-loop operating points and 9 of 12 open-loop points within an illustrative 1.0-3.0 mm root-side-width band. These simulation results support using the framework to screen welding conditions and control strategies before experimental tests. Thermal-vacuum, targeted reduced-gravity, and robotic hardware-in-the-loop studies are needed to assess the model and controller.
Figures & tables
Fig. 1: Concept for in-space robotic welding. A servicer-mounted six-axis manipulator positions the welding end-effector at a joint on the target spacecraft. Spacecraft geometry is illustrative. The manipulator simulation uses the UR5 kinematic description and Unified Robot Description Format (URDF) meshes published by Universal Robots [ 4 ] .
Fig. 2: Proposed architecture for the ISPARK digital twin. The finite-volume weld-pool model generates training data offline for the Gaussian-process and neural-network surrogate layer; the surrogate then serves the robotic control loop online. Indicative costs annotated at right show the latency difference between the finite-volume model and the control-loop requirement. The sensing interface is drawn dotted because it is specified in this work but not tested with hardware.
Fig. 3: Surrogate accuracy and cost. (a) Mean absolute error normalised by output range in four repeats of five-fold cross-validation; bars show fold-to-fold standard deviation. (b) Ten boundary cases held out of the Latin hypercube; 92.5% of 80 responses lie within nominal 90% GP intervals and median error is 1.64% of the output range. (c) Measured cost per operating point: the median of all 70 attempted solver cases and deployed surrogate queries including scaling and predictive variance. The three-output control query gives a speed-up of 1.6×106 .
Fig. 4: Plan-view liquidus boundary at the assumed reference condition ( 10−6g , 10−3 Pa). The finite-volume contour is grey. An ellipse from surrogate length and width gives IoU=0.68 at the torch (dashed) and 0.95 after axial placement (solid), showing that position dominates the error. Blockiness reflects the 0.25×0.125 mm cells.
open loop
closed loop
Model root-side width (mm)
2.010–3.678
1.722–2.088
Stations within band
9 of 12
12 of 12
Peak excursion (mm)
0.678
0.000
Surrogate MAE (mm)
0.089
0.049
Surrogate max error (mm)
0.247
0.188
Surrogate bias (mm)
+0.077
+0.034
Table 1: Finite-volume assessment at 24 commanded operating points. Station counts are not length fractions: the stations are clustered at the corner and therefore over-weight it by construction.
Fig. 5: Model-based closed-loop study for an L-shaped seam in 1 mm 316L at the assumed microgravity/vacuum reference condition. (a) Illustrative servicer, target and UR5 cell; the robot uses the published kinematic parameters. (b, c) Simulated path coloured by the illustrative 1.0 – 3.0 mm root-side-width band. (d) Open- and closed-loop predictions. (e) Travel speed and commanded power-source setting. The two loops use the same trajectory; slowing from 12.0 to 8.94 mm s -1 raises heat input per unit length by 34% at fixed setting, whereas control limits the rise to 5.5% and keeps the surrogate prediction inside the band.
Fig. 6: Model-based sensitivity of the GP surrogate. (a) Length scales fitted by automatic relevance determination after input standardisation; larger values indicate weaker response within the sampled domain. Values at the upper bound are marked. (b) Mean predicted output range obtained by varying one input over 40 levels across its sampled range and averaging over 300 uniformly drawn background points (random seed 0), normalised by the corresponding output range. The pressure input represents effective local pressure, not chamber pressure. The diagnostic describes the reduced model and should not be interpreted as experimental process sensitivity.
Fig. 7: Finite-volume temperature field for the reference calculation. (a) Top-surface plan view; (b) longitudinal section; and (c) transverse section at the maximum surface width. The solidus and liquidus contours distinguish the mushy region from the liquid weld pool. Dimensions annotate the liquidus contour: pool length 5.2 mm, top-surface width 2.50 mm and root-side width 2.00 mm. The depth is limited by the 1 mm sheet thickness. These are modelled liquidus-contour dimensions, not measurements of a solidified weld bead.
Robotic welding is widely used in industrial manufacturing, where the welding torch is often connected to the generator through an external umbilical. With the increasing deployment of lightweight and collaborative robots, the dynamic influence of this umbilical can significantly affect the robot motion and the actuation forces. This paper proposes a constrained multibody dynamic model of a welding umbilical, represented as a serial chain of rigid bodies interconnected by passive joints with elastic and dissipative effects. Prescribed motions at the distal anchor point are introduced through holonomic kinematic constraints. The equations of motion are reduced by projecting the dynamics onto the subspace of admissible velocities, yielding an efficient formulation free of Lagrange multipliers. The reaction wrench exerted by the umbilical on the robot is explicitly recovered. A planar case study illustrates the approach.
Nicolas Gautier, Yves Guillermit, Mathieu Porez +2
LS2N - équipe RoMas, LS2N · Weez-U Welding, 41-43 Quai de Malakoff, 44 000 Nantes, France · Nantes Université, École Centrale Nantes, Institut Mines Télécom Atlantique, CNRS, LS2N, UMR 6004, F-44000 Nantes, France +1
Real-time weld-pool perception is critical for closed-loop control in laser wire-feed welding, where sensing, computation, and actuator response introduce unavoidable delay. This paper presents a physics-guided spatiotemporal state space network for lookahead weld-pool segmentation. The model uses historical coaxial grayscale images, welding process parameters, and aligned wire-state electrical signals to predict the future semantic layout of three physically meaningful regions: keyhole, wire, and molten pool. It combines a visual encoder, process- and sensor-conditioned feature normalization, patch-level temporal state space modeling, horizon-conditioned latent prediction, dense future feature prediction, and a motion-aware mask decoder. Auxiliary signed-distance-function supervision, temporal consistency, feature distillation, and fine-grained keyhole losses further constrain the predicted geometry and local motion. Experiments on a 43-sequence laser welding dataset show that the proposed WeldMamba reaches 74.63% mIoU at a 500 ms lookahead. Ablation studies further show that temporal history, patch-level state space modeling, and keyhole motion awareness are the main contributors to robust future segmentation.
Sen Li, Haichao Cui, Changhao Yin +4
Shanghai Key Laboratory of Materials Laser Processing and Modification, School of Materials Science and Engineering, Shanghai Jiao Tong University, Shanghai 200240, PR China
Digital twins promise to enhance robotic manipulation by maintaining a consistent link between real-world perception and simulation. However, most existing systems struggle with the lack of a unified model, complex dynamic interactions, and the real-to-sim gap, which limits downstream applications such as model predictive control. Thus, we propose GaussTwin, a real-time digital twin that combines position-based dynamics with discrete Cosserat rod formulations for physically grounded simulation, and Gaussian splatting for efficient rendering and visual correction. By anchoring Gaussians to physical primitives and enforcing coherent SE(3) updates driven by photometric error and segmentation masks, GaussTwin achieves stable prediction-correction while preserving physical fidelity. Through experiments in both simulation and on a Franka Research 3 platform, we show that GaussTwin consistently improves tracking accuracy and robustness compared to shape-matching and rigid-only baselines, while also enabling downstream tasks such as push-based planning. These results highlight GaussTwin as a step toward unified, physically meaningful digital twins that can support closed-loop robotic interaction and learning.
Yichen Cai, Paul Jansonnie, Cristiana de Farias +2
Intelligent Autonomous Systems Lab, Technical University of Darmstadt, Germany. · NAVER LABS Europe. · Hessian.AI, Germany. +2