Enhancing Industrial Training through Situated Visualization in Augmented Reality: A Comparative Study of 2D vs. 3D Labels in Maintenance Tasks

Sammanfattning

Introduction: Industrial training is important for knowledge sharing, keeping workers safe, and guiding them to perform tasks successfully. Traditional training methods in industrial environments often lack real-time guidance and interactive feedback, making knowledge transfer challenging, which can be harder for novice workers who intend to gain practical skills. Augmented reality (AR) and situated visualization (SV), which present spatially relevant, in-depth instructions directly in the environment, offer a promising solution. However, there is a lack of validated, decision-ready guidance on how to design these visualization patterns to enhance training outcomes. By consulting with industry experts, this thesis presents an immersive tool deployed on head-mounted displays (HMDs) and explores how two different SV patterns -- two-dimensional (2D) vs.~three-dimensional (3D) labels -- affect user experience (UX) and task performance in an AR-based training scenario for machine maintenance. Research Question: The research question in this study is: ``How do different SV label designs (2D vs. 3D) affect UX and task performance in an AR-based training environment for industrial maintenance tasks?'' UX is assessed through perceived usability, cognitive and physical workload, and discomfort; task performance is measured by task completion time and memory recall. Method: An experimental research strategy compares two instructional designs—2D vs. 3D labels— within a virtual reality (VR)-simulated AR training environment. The training simulation is developed using Unity 3D and deployed on the Meta Quest Pro HMDs. The user study involves 24 convenience-sampled participants, including students and acquaintances representing novice trainees. Each SV condition is tested by 12 participants. Quantitative data are collected through standardized questionnaires (SUS, NASA-TLX, and SSQ), task completion time measurements, and a delayed memory recall questionnaire. The results are analyzed and discussed to evaluate how different SV designs impact usability, workload, discomfort, task completion time, and memory recall. Results: The results show that both SV conditions achieve high usability in AR-based industrial training. Although differences between conditions are not statistically significant, the 3D condition shows shorter task completion times, lower discomfort, and slightly lower workload, whereas the 2D condition shows a slight advantage in usability. In general, memory recall performance appears similar across both designs; however, conclusions remain limited due to several missing responses. Discussion: This study reveals that both SV designs effectively support AR-based industrial training by providing clear, spatially relevant guidance for the safety-critical Lock Out, Tag Out, Try Out (LoToTo) procedure. The primary finding is that there were no statistically significant differences between the 2D video labels and 3D ghost labels across all outcome measures. This suggests that both visualization methods were highly effective and achieved high usability scores. This shows that for SV training, displaying clear, spatially relevant guidance is more important than the specific dimensional format. Despite the lack of statistical significance, observable trends indicate designers can prioritize the 3D ghost labels when speed and user comfort are critical, and use 2D video labels when higher perceived usability and a familiar user interface are expected to be the priority. Limitations include student participants instead of real industrial trainees, the absence of a comparison to traditional training methods, and using a VR-simulated environment rather than deploying AR on the real factory floor. Future research should involve novice industrial workers, explore additional forms of SV beyond labels, include traditional training methods in comparisons, and aim to develop the project in AR to take advantage of the tangible aspect of the actual machines. Insights from this work contribute to designing AR tools that may enhance learning efficiency by improving task performance and UX in industrial training settings.