logo


Jul-2020

Evolution of a digital twin Part 2: Use of the digital twin

A two-part article provides a step-by-step introduction to the concept, construction and application of a digital twin for the lifecycle of an ethylene plant’s steam cracker.

OTMAR LORENZ, BERND-MARKUS PFEIFFER, CHRIS LEINGANG and MATHIAS OPPELT
Siemens

Viewed : 2437


Article Summary

Part 1 of this article (see PTQ Q1 2020) describes the development of a digital twin for application over the entire lifecycle of a process plant. Here, Part 2 of the article describes how, after the implementation phase is completed, it is possible to work with the digital twin to answer specified tasks in the engineering and operational area of the plant.

Engineering
Typical segments of engineering where the digital twin can be used are in the design of basic engineering of process equipment and the automation system, virtual commissioning of the control system, and the training before start-up of a new plant or reconfiguration of an existing plant.

Design
The goal of simulations in the design of a process engineering plant is the creation, verification, and refinement of the plant design. The focus is on considering the actual process. Controllers are only available in simplified form, if at all, as part of the process model. It is imperative that different process drafts can be compared with one another in order that the most suitable can be selected respectively. The accuracy of the simulation must be sufficiently good to be able to make process-related decisions correctly.

A static process simulation is sufficient for the design of plants in steady state continuous operation; a dynamic process simulation must be used for the simulation of start-up and shutdown processes and the transients between operating points. It may prove practical to combine models of different tools, either by exchanging models or via co-simulation.1

Virtual commissioning
The aim of virtual commissioning is to achieve a fully tested automation system wherever possible.2 The main focus is on testing the implemented PLC application software, developed uniquely for every system. For testing — for instance, signal routing, continuous function charts (CFC), sequential function charts (SFC), faceplate and pictures for operator station (OS), and alarms — a simulation model can be used, which operates the complete communication interface between automation and field and is connected to the real (hardware-in-the-loop) or emulated (software-in-the-loop) control hardware. It is imperative for both set-up scenarios that at least the communication behaviour of field devices (actuators and sensors) is replicated in the simulation model.

Replicating process behaviour (physical behaviour) will also prove practical for testing SFCs. This can be done, for example, with the simulation of a cold commissioning, in which the behaviour of the process is observed as long as only water is pumped through the system as a medium and no chemical reactions take place. Extremely detailed process models are required wherever the controller is to be parameterised. The connection of existing process models via co-simulation can be exceptionally advantageous in this regard.

At least for the hardware-in-the-loop configuration, the simulation system must be capable of supplying and processing signals within stipulated real time. Simulation models are also implemented as part of the control program on the automation hardware in a special software-in-the-loop configuration, eliminating the need for additional simulation tools.3 However, these advantages are offset by certain disadvantages. The control program is altered following testing, and simulation-specific functions such as a virtual time (faster or slower than real time), snapshots (saving model states) or even co-simulations may be difficult to attain with the resources of the automation system, if at all. Test cases which could be created automatically4 and automatically executed would be beneficial in ensuring the most efficient test possible.

Training (OTS)
The objective of training simulation is to prepare operators for their tasks as effectively as possible. This encompasses both interaction with the process control system (ideally on the basis of the original operating screens and programs), as well as familiarisation with the reaction of the process itself. Training for interaction with the process control system can be realised in accordance with the selected modelling depth based on the model which was created for virtual commissioning.

For training related to the process itself, it is necessary to model this in detail. Such models are thus also ideal as training for limit situations, start and stop procedures, and emergency scenarios. It is therefore essential that training scenarios can be created and adapted. In addition, it must be possible to assess, compare, and verify the performance of trained personnel.5 Moreover, particular attention must also be afforded to the didactic concept when devising the scenarios.6

Plant operation
Typical segments of plant operation where the digital twin can be used are in the design of virtual sensors, advanced process control, optimisation, and maintenance systems. Various aspects of application are described in more detail in the following chapter.

Soft sensor
Soft sensors represent an important application of a digital twin during the operation phase. A soft sensor estimates an unknown process variable based on a model of the process and other available measured variables. Common examples include the Luenberger state observer7 or the Kalman filter,8 which are based on dynamic process models in the form of differential or difference equations. As all variables are known in the simulation model, the variables being estimated can be obtained directly.

Reverting back to the digital twin of the system will ensure that a model based soft sensor does not have to be modelled anew for each application. A dynamic process simulation which is already available must be analysed and, where necessary, the sub-model separated for the process section for which a soft sensor is required. It is then only necessary to parameterise and to validate the soft sensor algorithm using process data.

The effort afforded for implementation of a soft sensor is worthwhile if the estimated variable is essential for process control. Estimated variables can be applied for monitoring tasks in which the exothermic reaction is estimated, for example, and monitored for a maximum permissible value to avoid unfavourable or dangerous process states.9 Direct control of estimated variables is also possible. Thus, in the example given, the yield can be measured, but only after several steps of the procedure have been executed. The resultant dead time, which is many orders of magnitude greater than the actual process dynamics, renders direct control of the measured yield impossible. The estimated yield at the output of the cracking furnace, however, is provided free of dead time via the soft sensor and may thus be used for direct control.

APC
All higher level control procedures which go beyond standard single-loop PID controllers come under the APC keyword. In view of the task definition for a multivariable control on the steam cracker, model based predictive control (MPC) seems to be the most appealing option.10 All predictive controllers are based on the basic principle of internal model control (IMC). A dynamic model of the controlled system is part of the controller and is used during runtime to predict future process behaviour in a defined prediction horizon. The model knowledge of the digital twin can be used as a basis for the process model of predictive controllers.


Add your rating:

Current Rating: 4


Your rate: