AI Case Study
Shell plans to automate information extraction from internal documents to collate answers to operational problems using Maana's Knowledge Platform
Shell intends to implement Maana's knowledge management tools to compile and analyse a variety of company data in order to address specific problems, such as engineering research and health and safety incidents. Unstructured data in internal company documents is analysed and natural language processing is used to extract information.
Oil And Gas
"The Maana Knowledge Graph is specifically for industrial applications and is notably capable of capturing and containing not just data but also people’s expertise and their understanding of a particular business process or decision flow. Knowledge graphs are of great interest to Shell because they offer a way to represent, store and join data, and then to model them
easily and quickly. Within Shell, we have masses of data presented in different formats and, perhaps more importantly, recorded within different business contexts. Maana works with these people, the subject matter experts, to create a picture of what a business wants to achieve from the decision-making perspective
and then to break it down into discrete problemquestions, as we call them. These form the basis of the series of related business models, the Knowledge Models, that go to make up the Maana Knowledge Graph. Maana’s artificial intelligence algorithms are designed to help create Knowledge Models quickly. DocAssist is worth a particular mention, as it extracts unstructured information from PDF documents, for example, technical papers, reports or even emails." For the Knowledge Graphs, Maana uses natural language processing.
"Shell has two projects close to actually implementing the Maana Knowledge Graphs. The first is corrosion-related and concerns the
impact of crude oil selection on refinery equipment integrity. We are confident that a newly developed knowledge application will help engineers to understand more fully the corrosion risks associated with specific crudes. Its ultimate aim is to reduce maintenance costs and unplanned downtime. The second is a health, safety and environmental risk application designed to help Shell identify operational incidents that could have had a worse
outcome than they did. The aim is to improve our understanding of what really went wrong in these cases and thus prevent recurrences. This has involved aggregating and analysing information about health, safety and environmental incidents, as well as mitigation measures, in the company’s primary health, safety and environmental database and creating an algorithm that accurately predicts incident outcomes. Analysing the sheer volume of fragmented data, both structured and unstructured,
has been a big challenge. Traditional statistical tools and techniques are unable to help us solve this kind of problem. In addition, there are several other projects at an earlier stage of development. A particularly exciting and challenging one aims to improve decision making in the exploration area through
better understanding of reservoir basin features. Automatically extracting knowledge from scientific papers will feature heavily here."
"The Maana Knowledge Platform is one of the enabling technologies that can support Shell's vision of joining up the data from different parts of Shell in order to gain deeper insights into the broader enterprise. Decisions taken in one part of the business clearly affect others. A tool like this should help us to make improved business decisions by showing how any single decision
affects not just that particular part of the business but also wider elements of Shell’s value chain, thus creating greater overall business value... several successful pilot trials have been carried out using the Maana Knowledge Platform all under the stewardship of the Technical and Competitive IT (TaCIT) division of Shell Projects & Technology."
Maana claims its "Knowledge Assistants [algorithms] help customers to create models 3–10 times faster than when using any other technology".
Structured and unstructured, e.g. technical papers, reports, emails