Challenge

Snam is a leading European operator in gas infrastructure. They engaged Eng to automate the extraction of relevant data from various technical drawings (P&ID — Pipeline & Instrumentation Diagrams) provided by the designers of individual regulation and metering plants (ReMI), aiming to streamline their project and plant modification validation process.

The project involved thousands of different P&IDs, requiring advanced engineering expertise and significant time for analysis. Additionally, each plant follows its own specific design policy.

Approach

We adopted a gradual and iterative approach, starting with a limited set of plants and then expanding the scope of the solution to cover the entire fleet of meters.

To enable the extraction of relevant information from various technical plant drawings, we combined multiple algorithms based on Generative AI, Machine Learning, and Image Analysis.

Solution

We developed a system that, by integrating custom models tailored to the client based on relational transformers and multimodal large language models (LLMs), is capable of analyzing technical plant drawings and extracting their key features into a relational database.

Given a .pdf file containing the plant’s documentation as input, the solution uses Machine Learning algorithms to extract project data and a list of components, which are validated through Generative AI models.

Using a custom advanced Image Analysis model, the solution automatically recognizes the symbols present in the technical drawings and determines whether they are connected. Based on the detected symbols and relationships, a probabilistic system reconstructs the traces and lines that compose the individual regulation and metering plant.