Project R-17014

Spatially Anchored Diagnostic Knowledge: From Capturing Expert Demonstrations to Multimodal Guidance (Research)

Effective process diagnostics in high-variation industrial settings depends on experience-based judgement; knowing which checks are most discriminating, how to interpret ambiguous cues, and when evidence is sufficient to act. This expertise is rarely consolidated: dispersed across documentation and a few experienced technicians, it remains inaccessible for novices and experts alike. Existing approaches, from fault ontologies to LLM-assisted methods, fall short, are brittle under incomplete documentation and unable to capture the spatial reasoning underpinning real diagnostic work. This PhD pursues three objectives. RO1 develops a knowledge architecture in which hypotheses, evidence, and micro-tests are anchored to 3D spatial representations, making geometry, connectivity, and accessibility computable constraints that guide diagnostic reasoning. RO2 develops in-situ elicitation procedures that capture expert reasoning at the moment it is exercised, recovering tacit heuristics through targeted prompts during demonstrations rather than post-hoc interviews. RO3 investigates an interactive guidance paradigm that bootstraps diagnostic reasoning from vague user input, iteratively narrows the hypothesis space through discriminating micro-tests, and enables knowledge transfer across system variants. Together, these objectives aim to make diagnostic expertise more structured, capturable, and transferable, reducing barriers for technicians and industrial downtime.

01 November 2026 - 31 October 2030