Graph based reasoning for scaling energy audits to many customers.

In a drive to reduce the carbon footprint of their customers’ Grundfos is revolutionizing how they engage on efficiency savings of their cooling systems. Initial communication with the customers about cooling systems needs to capture complex system information, yet be simple and relevant for each of the many customer profiles to maximize retention. The solution has to work with a wide variety of engineering systems and many stakeholders within the client. Adopting traditional forms-based methods would have resulted in a complex, costly and error prone sales process where the most likely result was customer drop out. We have developed a process built on semantic reasoning that guides the process, asks the next best questions and simplifies the whole process for the users.

We will begin by setting the context and impact that efficiencies in cooling systems needed to execute on CO2 reduction and meet the 13th Sustainable Development Goal 13. We will give a high-level description of a cooling system control service that can both make an impact and is lead to a profitable outcome for both the consumers and suppliers. The app prototype has the goal of answering a simple business question: “will the energy savings method that the Grundfos team is deploying fit the cooling system of interest and yield sufficient energy savings to justify a partnership?”. Normally this question can only be answered by Grundfos experts performing interviews with system owners and facility managers – but the education of such Grundfos experts is a bottleneck for scaling the business.

The complexity of understanding the cooling system of interest with the energy saving in mind is sketched out and modelled as a graph. This avoids burying business logic in the application logic. By building a  knowledge graph for each system, the seed for a digital twin of the cooling system in a machine-readable format is captured.

This domain based knowledge graph combined with the answers iteratively given by users are reasoned over using RDFox. This allows the system to determine the next best question to ask. Within the talk we will demonstrate the system in action, show the architecture and user experience made possible through the application of reasoning and graph supported questioning. Whilst what is presented here will be focused on an engineering system we believe the same guided questioning approach has the potential to be applied to many applications.