Enhancing Asset Data Quality: Automating P&ID Tag Extraction

We are working on making our contextualization efforts more efficient. One of our key initiatives is to extract all tags from P&IDs to build a clean dataset that we can then match against asset hierarchies, helping us identify gaps on both sides. Additionally, we aim to minimize the need to revisit P&IDs whenever new patterns emerge that don’t align with the current contextualization process.

For our initial approach, we tested extracting all tags from P&IDs to create a comprehensive list. We are aware that Cognite provides an API feature to extract tags based on predefined entity patterns. However, we are looking for a solution that allows us to extract all tags—without requiring predefined patterns.

We have tested multimodal LLMs for this purpose, and they have shown promising results. Our questions are:

• Does Cognite offer any built-in functionality that supports extracting all tags from P&IDs without predefined patterns?
• If not, is there a cost-efficient approach to achieving this without heavily relying on third-party multimodal models? Specifically, if Cognite already integrates Azure services for this task, could we explore an alternative configuration to optimize cost and efficiency?

Additionally, we came across the discussion Extracting Asset Hierarchy from PDF P&IDs in Cognite Data Fusion and are curious if any similar approaches or insights from that topic could be applied to our challenge.

We would appreciate any insights into existing Cognite capabilities or recommended approaches to achieve our goal.