AI-Assisted Visual Mapping : of Industrial Facilities for Maintenance Management

Sammanfattning

Industrial maintenance workflows rely on Computerized Maintenance Management Systems (CMMS) to manage structured asset data. However, spatial representations of facilities are often maintained separately as 2D visual layouts, with limited support for richer 3D representations. Mapping and updating these visual structures is largely a manual process, making it time-consuming, difficult to maintain, and weakly integrated with hierarchical asset data. This thesis evaluates an AI-assisted approach for identifying functional elements in 2D industrial facility layouts, estimating their spatial placement, and integrating the results into CMMS workflows. The work combines two parts: an empirical evaluation of state-of-the-art multimodal AI-based systems and a proof-of-concept prototype that connects AI-generated layout interpretations to an interactive 3D environment. The empirical evaluation showed consistently high semantic correctness when functional elements were identified, indicating that multimodal AI systems can interpret visual and textual layout cues reliably. Spatial placement was less reliable, with common errors including imprecise localization, systematic shifts, and oversized bounding regions. These results suggest that current models are useful for layout interpretation, but lack the geometric precision required for fully automated mapping. A user evaluation with industry experts showed that AI-assisted features, such as pre-generated placement suggestions, can reduce perceived manual effort while preserving user control. Their usefulness depended on interpretable and easily adjustable outputs, highlighting the importance of human-in-the-loop interaction design. Overall, the thesis shows that multimodal AI can provide support in industrial layout mapping, but is not yet sufficient for full automation. Its main value lies in augmenting human workflows rather than replacing manual processes. The thesis contributes an empirical evaluation of AI capabilities in this context and design insights for integrating AI-assisted visual mapping and 3D spatial support into CMMS systems.