System engineering, plant maintenance, and service organizations spend a lot of time finding all the necessary information to plan services, upgrades, and enhancements to an existing plant, factory, and other large systems.
Finding the full set of documents and photos is time-consuming even if only one pump needs to be replaced, for example. When engineers and service persons spend dozens of hours every week to manually find connections on documents, that results in millions of euros wasted every year.
The hidden value
There is a lot of meta information available in photos, drawings, and PDF documents but it is not easily accessible. This information is typically not linked, tagged, or structured to be searchable and browsable. For example, there are no connections from part numbers to document paragraphs and sentences. Neither are there summary descriptions from multiple documents as browsable database records with related photos.
Engineering needs a system that organizes unstructured data into structured information automatically. This data then needs to be available for intelligent search and browse directly in engineering tools. An engineering knowledgebase is the answer to the need to have faster findings.
Machine Learning (ML) can be used to find engineering information and create connections to the documents and images.
Partnership to dig out the value
All3D Oy and Koivu Solutions Oy formed a partnership and developed an ML based system to increase engineering data creation and linking productivity.
All3D has deep domain knowledge of building digital twins. One of their key challenges is that the manual search-and-connect process for engineering information takes a lot of time and energy. Having an AI to assist in this process is speeding-up the value delivery significantly.
Koivu developed a state-of-the-art ML system that recognizes engineering information from images taken from a plant, shop floor, ship’s engine room, etc. For PDF documents written by design engineers, Koivu developed an NLP*-based system that automatically creates links between part numbers, sentences, paragraphs, and documents. Then a searchable knowledgebase is built of this information.
*) NLP stands for Natural Language Processing, so a method of how systems can understand written text. In this case, the result of this is a triplet database, so a browsable database of subject-predicate-object sentences associated with meta-data.