An industrial knowledge graph is a record of every piece of equipment in a plant and how each connects to the others. It is the skeleton that a digital twin hangs its data on. A knowledge graph is a way of storing facts as items and the links between them. The general idea is described in enterprise knowledge graph.
In a plant, each pump, valve, pipe and sensor is an item. Each link has a plain meaning: "feeds", "measures", "isolates" or "is part of". For example, a link can say that a pressure sensor "measures" a reactor, or that a valve "isolates" a pump. The graph is stored in a graph database, which is built to follow such links quickly.
The graph answers questions that a list of equipment cannot. Consider a plant engineer who must stop one pump for repair. She asks: "What is affected if this pump stops?" The graph follows the "feeds" links from the pump, one step after another, and lists everything downstream. A list of equipment would only show that the pump exists. In a plant with tens of thousands of items, finding the answer by reading drawings can take a day. The graph gives it in seconds.
The graph is useful in other ways too. When a new sensor is installed, it is linked to the equipment it measures, so its readings appear on the right item in the plant digital twin. A safety engineer can ask which valves must be closed to isolate a section before work begins. And an alert about one machine can show at once which other machines are connected to it.
The graph is built from drawings made in plant design software and from records in an industrial data platform. This is slow work, because the names in drawings and records rarely match. It must also be kept up to date. When a pipe is rerouted, the graph must be changed, or its answers will be wrong.