FlowExtract: Procedural Knowledge Extraction from Maintenance Flowcharts
Summary: arXiv:2604.06770v1 Announce Type: cross
Abstract
Maintenance procedures in manufacturing facilities are often documented as flowcharts in static PDFs or scanned images. They encode procedural knowledge essential for asset lifecycle management, yet remain inaccessible to modern operator support systems. Vision-language models, the dominant paradigm for image understanding, struggle to reconstruct connection topology from such diagrams. We present FlowExtract, a pipeline for extracting directed graphs from ISO 5807-standardized flowcharts. The system separates element detection from connectivity reconstruction, using YOLOv8 and EasyOCR for standard domain-aligned node detection and text extraction, combined with a novel edge detection method that analyzes arrowhead orientations and traces connecting lines backward to source nodes. Evaluated on industrial troubleshooting guides, FlowExtract achieves very high node detection and substantially outperforms vision-language model baselines on edge extraction, offering organizations a practical path toward queryable procedural knowledge representations. The implementation is available at https://github.com/guille-gil/FlowExtract.
Introduction
In modern manufacturing environments, the efficiency of maintenance procedures is crucial for ensuring asset reliability and minimizing downtime. However, the documentation of these procedures often exists in a form that is not easily accessible for automated systems. Flowcharts, which provide a visual representation of processes, are frequently stored as static PDFs or scanned images, making it difficult for contemporary operator support systems to leverage this valuable procedural knowledge.
Challenges with Current Paradigms
Current vision-language models, while advanced, encounter significant limitations when it comes to interpreting flowcharts. The key issues include:
- Connection Topology Reconstruction: These models struggle to accurately identify how different components of a flowchart are interconnected.
- Element Detection: Existing solutions often fail to achieve high accuracy in detecting nodes and text within flowchart diagrams.
Introducing FlowExtract
FlowExtract addresses these challenges head-on. This innovative pipeline separates the tasks of element detection and connectivity reconstruction, allowing for greater accuracy and efficiency. The key components of FlowExtract include:
- YOLOv8: Utilized for high-performance object detection, this model is tailored to identify nodes within flowcharts.
- EasyOCR: This text extraction tool complements YOLOv8 by accurately retrieving textual information from detected nodes.
- Novel Edge Detection Method: FlowExtract employs a unique approach to analyze arrowhead orientations and trace connecting lines back to source nodes, ensuring a comprehensive understanding of flowchart connectivity.
Evaluation and Results
FlowExtract was rigorously evaluated using industrial troubleshooting guides, demonstrating exceptional performance in node detection. The results indicate that it significantly outperforms baseline vision-language models in edge extraction tasks. This advancement not only enhances the accessibility of procedural knowledge but also provides organizations with a practical solution for converting flowcharts into queryable directed graphs.
Conclusion
FlowExtract represents a significant leap forward in the extraction of procedural knowledge from maintenance flowcharts. By enabling the conversion of static diagrams into dynamic, queryable formats, it paves the way for improved asset management and operational efficiency in manufacturing facilities. For those interested in exploring this technology further, the implementation is available at https://github.com/guille-gil/FlowExtract.