A Linear Topology-Aware Alignment Framework for Belt Conveyor Maintenance Knowledge Graph

To address spatial confusion and semantic collapse in maintenance knowledge graph construction for long-distance belt conveyor systems, this study proposes a linear topology-aware alignment framework driven by large language models. The framework constructs a Physical–Space–Fault–Rule ontology and introduces a normalized linear topology mapping mechanism to assign unique spatial anchors to highly homogeneous components. For noisy and colloquial maintenance logs, a Condition–Constraint–Action–Space (CCAS) extraction mechanism is developed to reconstruct implicit fault causal chains and spatial information from unstructured text. To prevent erroneous merging of components with similar semantic descriptions but different physical locations, the LTA-Alignment algorithm integrates Gaussian topological penalties with LLM-based grey-zone arbitration. A structured gold-standard annotation protocol is further established using maintenance logs, BOM records, spatial anchors, and closed-loop repair evidence. Experiments over three predefined runs show that CCAS achieves a global F1-score of 94.10 ± 0.56%, exceeding T5 by 6.20 percentage points. LTA-Alignment achieves a Merge F1-score of 95.75 ± 0.53% and an entity cluster purity of 96.70 ± 0.53%, exceeding SCSA by 3.36 and 3.90 percentage points, respectively. In downstream fault traceability, the dynamic spatiotemporal knowledge graph achieves an Accuracy@1 of 92.78 ± 0.36%, 10.16 percentage points higher than R-GCN, together with an estimated 68.50 ± 0.73% reduction in rule-based simulated troubleshooting time relative to Keyword Retrieval. The results demonstrate that combining normalized spatial topology with LLM reasoning improves entity disambiguation, fault-causal reconstruction, and traceability performance under the evaluated non-branching belt conveyor maintenance setting.