WELDAR: Augmenting Live Hands-On Training with In-Situ Guidance for Novice Learners

Paper Title

WELDAR: Augmenting Live Hands-On Training with In-Situ Guidance for Novice Learners

Publication Info

• Topic area: Augmented Reality (AR) systems for vocational skill training, focusing on live welding.
• Keywords: Augmented Reality, welding training, in-situ guidance, skill transfer, vocational education, real-time feedback, embodied learning, XR systems, usability, MIG welding.

Background and Problem

• Problem / challenge: Traditional welding training relies on post-hoc feedback and is limited by instructor availability, while video-based instruction lacks real-time feedback. Existing XR systems focus on simulations, which fail to replicate the sensory and embodied demands of live welding, leading to skill transfer gaps.
• Significance: Welding is a critical vocational skill, but training scalability is constrained by instructor shortages and the need for hands-on, real-time guidance. Addressing these gaps can improve training efficiency and skill retention.
• Motivation and related work: Prior XR systems, including VR and AR simulators, have shown promise in early skill acquisition but lack predictive validity for live welding. This paper builds on the need for in-situ AR systems that provide real-time, sensory-aligned feedback during live welding.

Solution

• Proposed approach: WeldAR, an AR system integrated into a welding helmet, provides real-time, in-situ guidance for novice welders on four key parameters: Contact Tip to Work Distance (CTWD), travel angle, work angle, and travel speed.
• Novelty:
  1. Development of WeldAR, a live AR system tailored for MIG welding, including hardware, sensing, and instructional modules.
  2. Empirical evidence showing AR's superiority over video instruction in improving welding performance and skill transfer.
  3. Insights into the usability, workload, and experiential factors influencing AR effectiveness in vocational training.
  4. Design implications for AR systems supporting embodied learning in high-risk, hands-on domains.
• Procedure and key techniques:
  • Hardware: Meta Quest 3 headset integrated into a custom welding helmet, torch-mounted controller, and calibration station.
  • Software: Real-time feedback on four welding parameters, scaffolded lessons, and post-lesson performance summaries.
  • Study: Within-subjects crossover study with 24 novices comparing AR and video instruction, including assisted and unassisted welding tasks.

Results

• Concrete findings:
  • AR training significantly improved performance on travel speed and work angle, with composite z-score deviations lower than video (F1,22=23.40, p<.001, η²=.52).
  • AR-first participants maintained stable performance when switching to video, indicating better skill transfer.
  • AR was rated higher in perceived usefulness (M=6.54 vs. 4.87, p<.001) and engagement (M=6.62 vs. 4.21, p<.001).
• Advantage over baselines:
  • AR outperformed video in both assisted and unassisted tasks, with significant improvements in skill retention and consistency.
  • AR-first participants showed no performance decay when feedback was removed, unlike video-first participants.
• Experiments / evaluation:
  • Participants: 24 novices (18-35 years old) with no prior welding experience.
  • Design: Within-subjects crossover study comparing AR and video instruction.
  • Metrics: Composite z-scores for welding parameters, NASA-TLX workload ratings, and TAM-based usability surveys.
• Limitations and future work:
  • Hardware issues: Controller drift, headset weight, and visual resolution constraints.
  • Study design: Participant fatigue and limited time for reflection.
  • Future directions: Improved tracking fidelity, adaptive feedback fading, VR comparisons, and longer-term studies on skill retention.

Summary

WeldAR is an in-situ AR system designed to enhance live welding training by providing real-time feedback on key welding parameters. A user study with 24 novices demonstrated that AR significantly outperformed video instruction in improving welding performance, skill retention, and user engagement. AR training facilitated skill transfer by fostering embodied learning and muscle memory, though challenges such as hardware ergonomics and cognitive overload were noted. This work highlights the potential of AR to scale vocational training and suggests design strategies for supporting embodied skill acquisition in high-risk domains. Future research will address hardware limitations, explore adaptive feedback mechanisms, and evaluate long-term learning outcomes.