(79c) Architecture and Implementation of a New Generation of Software for Consistent Life-Cycle Digital Twin Models of Process Plants

Current digital twin models of process industry plants are based on different representation of plant topology (e.g. detailed real-time optimization models includes all pieces of equipment, while production scheduling or production planning models often omit many of them, such a pumps or heat exchangers).

This dichotomy is caused by the need for acceptable model execution times. For example, multi-time period planning models are based on a plant network that is as simple as possible and equipment models also are linear or close to linear. Consequences of disparate models being used for different decision making (planning, scheduling, real-time optimization, and control) is that the answers obtained in one decision making step are often inconsistent with answers obtained in another decision-making step (based on a different plant model).

We introduce a detailed plant flow diagram (PFD) as the basis for different incarnations of a plant model. Each incarnation of the plant model is based on a different level of abstraction of nodes and streams (e.g. mass-balances only, or mass and energy balances, etc.) and all incarnations are based on the same plant topology. This makes it possible to use a solution from a higher level of abstraction (e.g. material balance planning model) as a starting point for a solution at another level (e.g. scheduling or RTO). It also enables us to model the plant by nodes at different levels of abstraction in different time periods (more detailed models in near-horizon time periods and simpler models in far-horizon time periods).

Multi-phase composite algorithms that successively solve mass and energy balances, with update of local approximations of thermodynamic properties are presented.

The new architecture is implemented in Python; it uses PYOMO as an underlying framework for aggregating and solving all equations comprising a plant model. Availability of machine learning and AI tools in Python this software implementation particularly suitable for inclusion of data-driven and hybrid node models. Browser-based GUI utilizes noSQL database (MongoDB) to store plant flow diagram representation and attributes describing nodes and streams. Novel, more accurate planning and scheduling models made possible by this paradigm are described not only by material balances but also by energy balances. Integration of plant control model (e.g. data-driven MPC model) with mass and energy balances enables rapid solution of dynamic RTO via multi-period model structure which can also cover the scheduling horizon. Since process flows in all incarnations of the plant models are based on the same PFD, there is no need to translate / re-interpret e.g. results from a planning model to be able to use them in a scheduling model.

Rapid execution makes it possible to integrate long-term planning with plant scheduling without having to build different representation of the plant for each of these decision making processes. An example of modelling a blue hydrogen plant demonstrates the versatility of the new software architecture and the speed of computation.