Knowledge-Driven Validation of Procedural Steps for XR-Based Industrial Assembly

Sammanfattning

Extended Reality systems and AI-based recognition methods can capture operator actions, object interactions, and spatial relations during industrial assembly tasks, but these observations do not automatically become explicit procedural knowledge. Upstream AI-generated step descriptions and perception-derived predicates may remain descriptive, omitting conditions, dependencies, constraints, compatibility assumptions, and reasons for step validity.This thesis investigates how ordered procedural step hypotheses can be represented and validated through a knowledge-driven reasoning layer for XR-based industrial assembly. The proposed approach represents step-level evidence as symbolic predicates, infers procedural constraints using rules and domain knowledge, and validates ordered step hypotheses under incomplete and confidence-weighted evidence.The implementation connects upstream step and component artifacts to validation records, explanation traces, and a procedural reasoning graph.The evaluation shows that, under assumed symbolic input conditions, the pipeline produces the required artifacts, infers procedural constraints, responds consistently to missing, low-confidence, or incompatible evidence, and exposes rule-based validation decisions through graph-based traces. The results indicate that the reasoning layer can ground AI-derived procedural artifacts in explicit symbolic evidence and connect observation-derived procedure representations to context-aware operator assistance.By making validation decisions inspectable, the approach supports more reviewable integration of XR and AI systems in industrial settings.