2.2 Design of Digital Twins as Control Strategy
Development Tools
To utilise a Digital Twin for the development of both conventional (e.g. single loop
PID control) and advanced control (e.g. multivariable controllers, model predictive
control), it must fulfil specific requirements that have to be considered during the
design process of the Digital Twin. According to Hass [11], desirable characteristics
of a functionally useful Digital Twin include realistic simulation of the biological,
physical and chemical processes, accurate representation of automation and control
actions and a GUI with a similar “look and feel” to that of the real plant [11]. Mathematical models used in Digital Twin development are classified broadly as mechanistic, non-mechanistic or hybrid models [9, 10, 12]. In this context, a model refers
to a mathematical representation of certain aspects of a real-world object or phenomenon. Non-mechanistic models use sets of experimental data to represent
observed phenomena by fitting parameters based on the available datasets. Mechanistic models seek to represent experimental observations based on the underlying
biological, chemical, and physical mechanisms occurring in the system. Mechanistic
models offer excellent predictive capabilities beyond the original experimental
conditions used for model development. By contrast, non-mechanistic models only
offer very restricted predictive capabilities [2, 9–12]. Mathematical modelling for a
Digital Twin involves several key steps. The first step is a definition of the process
using appropriate diagrams and charts. A process flow diagram and a piping and
instrumentation diagram (P&ID) are excellent starting points for system definition
[10, 13, 14]. Ideally, verbal process description and expected modelling targets
including levels of model accuracy are specified at this stage. Following system
definition, appropriate mathematical models that sufficiently describe the physical,
biological, and chemical processes in the system are formulated based on literature
research [9, 14]. To structure the process model, it has been suggested to divide the
model into smaller sub-models. One approach is the shell model introduced by
Blesgen et al. [15, 16] and extended by Hass et al. [17]. In this case, the overall
mathematical model of the Digital Twin is divided into a biological sub-model,
physico-chemical sub-model, a reactor sub-model, a plant and peripheral sub-model
as well as a control and automation sub-model (see also Chapter: Moser, Appl,
Brüning, Hass “Mechanistic Mathematical Models as a Basis of Digital Twins for
process optimization”, which is also in this book series). Depending on the requirements of the Digital Twin, the shell model can be extended or reduced in complexity.
2.2.1 Software Tools for the Design of Digital Twins
Further steps in Digital Twin development include model implementation using
suitable tools, model parameterisation and finally model validation using experimental data. Several modelling tools for the development of Digital Twins are
readily available and easy to use, but they do not provide the flexibility and
68
C. Appl et al.
Development Tools
To utilise a Digital Twin for the development of both conventional (e.g. single loop
PID control) and advanced control (e.g. multivariable controllers, model predictive
control), it must fulfil specific requirements that have to be considered during the
design process of the Digital Twin. According to Hass [11], desirable characteristics
of a functionally useful Digital Twin include realistic simulation of the biological,
physical and chemical processes, accurate representation of automation and control
actions and a GUI with a similar “look and feel” to that of the real plant [11]. Mathematical models used in Digital Twin development are classified broadly as mechanistic, non-mechanistic or hybrid models [9, 10, 12]. In this context, a model refers
to a mathematical representation of certain aspects of a real-world object or phenomenon. Non-mechanistic models use sets of experimental data to represent
observed phenomena by fitting parameters based on the available datasets. Mechanistic models seek to represent experimental observations based on the underlying
biological, chemical, and physical mechanisms occurring in the system. Mechanistic
models offer excellent predictive capabilities beyond the original experimental
conditions used for model development. By contrast, non-mechanistic models only
offer very restricted predictive capabilities [2, 9–12]. Mathematical modelling for a
Digital Twin involves several key steps. The first step is a definition of the process
using appropriate diagrams and charts. A process flow diagram and a piping and
instrumentation diagram (P&ID) are excellent starting points for system definition
[10, 13, 14]. Ideally, verbal process description and expected modelling targets
including levels of model accuracy are specified at this stage. Following system
definition, appropriate mathematical models that sufficiently describe the physical,
biological, and chemical processes in the system are formulated based on literature
research [9, 14]. To structure the process model, it has been suggested to divide the
model into smaller sub-models. One approach is the shell model introduced by
Blesgen et al. [15, 16] and extended by Hass et al. [17]. In this case, the overall
mathematical model of the Digital Twin is divided into a biological sub-model,
physico-chemical sub-model, a reactor sub-model, a plant and peripheral sub-model
as well as a control and automation sub-model (see also Chapter: Moser, Appl,
Brüning, Hass “Mechanistic Mathematical Models as a Basis of Digital Twins for
process optimization”, which is also in this book series). Depending on the requirements of the Digital Twin, the shell model can be extended or reduced in complexity.
2.2.1 Software Tools for the Design of Digital Twins
Further steps in Digital Twin development include model implementation using
suitable tools, model parameterisation and finally model validation using experimental data. Several modelling tools for the development of Digital Twins are
readily available and easy to use, but they do not provide the flexibility and
68
C. Appl et al.