Forthcoming, with Monash Pilot Processes, Chemeca 2024

Presentation abstract

Any description of a chemical process system—manuals, standard operating procedures, inventory lists, software simulation—documents some understanding of how the system works. With traditional documentation methods, there is significant risk of a set of documents becoming inconsistent as modifications are made to the system or errata are discovered. Resolving this can be an administrative burden when working with text documents, but this is worsened if, for example, software needs to be modified to fix an underlying assumption. Identifying points of inconsistency can be difficult, and even trivial ones have the potential to cast doubt on the entire set of documents.

We present a digitalised approach to knowledge management which uses a knowledge graph as the single source of truth. The knowledge graph encapsulates the functions of the chemical processes of a system, their underlying physics, and their material realisation in the actual system; all other documents are derived from the graph using symbolic artificial intelligence (AI). We then describe a specific application of this approach in the development of operator training software for a membrane-based water treatment pilot plant (WTPP). The software simulates the behaviour of the WTPP’s processes in response to user actions—to mimic the rich possibility space of all states, and interaction between states, of the physical equipment. It provides instructions to guide users through an experimental procedure reactively, accomplished using finite automata. We discuss the following components of the software:

• Creating a knowledge graph representation of the WTPP extending the OntoCAPE domain ontology, following semantic web best practices;
• Simulation of membrane filtration processes using a mechanistic, mathematics-based model of process and water quality parameters, reasoning water flow paths from the knowledge graph;
• Authoring large and complex finite automata for a given experimental procedure by reasoning over the knowledge graph.

For effective knowledge management, we briefly discuss each component’s adaptability to changes to the WTPP or its operation. Prioritising correctness and observability as criteria, we identify classes of errors which can occur in each of these components and suggest how they can be diagnosed. Finally, we suggest how the approach can be applied for authoring more documents—such as standard operating procedures—to form a knowledge management system, retaining the correctness and observability demonstrated in the operator training software.