The effective operation and maintenance of facilities requires a comprehensive description of the assets that constitute them. The three components of the description are the asset register, bills of material and material catalog. However, many companies only manage to create these components partially because:

• The source data is difficult to find and extract since it is often siloed, trapped in scanned documents and represented with inconsistent terminology
• Structuring whatever is found and extracted is done in Excel although it is not optimal for this
• The time and cost of human participation to compensate for the above is high

This session presents the use of computer vision and natural languages processing technologies to find and extract the relevant data from engineering drawings, parts tables and data sheets, and then organise it in a knowledge graph structured in accordance with equipment classes. This will support better and faster decision making for maintenance planning and spares optimisation.

The technology stack for the solution is comprised of Python, TensorFlow and Neo4j software.