20th European Symposium on Computer Aided Process Engineering - ESCAPE20 S.

Pierucci and
G. Buzzi Ferraris (Editors)
2010 Elsevier B.V. All rights reserved.

Practical Aspects of Dynamic Simulation in Plant

Engineering
Gabriele Engl, Andreas Kroner, Martin Pottmann
Linde AG, Linde Engineering Division, Dr.-Carl-von-Linde-Str.6-14, Pullach, Germany
E-mail: gabriele.engl@linde-le.com, andreas.kroener@linde-le.com,
martin.pottmann@linde-le.com

Abstract
This paper illustrates the increasing trend to integrate dynamic simulation into the
workflow of a plant engineering and contracting company. While methods and tools for
dynamic simulation are well-established in academia, the industrial world still faces a lot
of challenges when it comes to their practical application. These challenges are discussed
by example of a carbon capture and storage (CCS) technology: an oxyfuel power plant.
Dynamic simulation is first applied to the individual plant components such as the Gas
Processing Unit (GPU) and the Air Separation Unit (ASU), where the early integration
into the plant engineering workflow plays a major role. In order to study the dynamic
behavior of the complete oxyfuel power plant in a second step, the models which are
implemented in different simulation tools are reduced for integration into a common
simulation platform.
Keywords: Dynamic simulation, carbon capture and storage, oxyfuel, model reduction

1. Introduction
Computer Aided Process Engineering (CAPE) plays a key role in the industrial practice
of the Linde AG, Engineering Division. As a leading international engineering and
contracting company, Linde Engineering designs and builds turnkey process plants for a
wide variety of industrial users and applications: The chemical industries, air separation,
hydrogen and synthesis gas production, natural gas processing and more. Being able to
call on its own extensive process engineering know-how in the planning, project
development and construction of turnkey plants, Linde Engineering is also pursuing a
CAPE strategy based on a long tradition of internal know-how: High quality methods
and tools available from universities or commercial suppliers are combined with internal
developments to achieve optimal solutions. For process simulation and optimization,
Linde Engineering uses commercial process simulation tools as well as its in-house
process simulation program OPTISIM [1,2].
While steady-state simulation and optimization are well established in the engineering
workflow, the use of dynamic simulation is still restricted to a small number of experts
and selected projects - even though the methods and tools have been available for a long
time. However, there is an increasing trend to apply dynamic simulation, in particular for
new process developments such as in the area of CCS. This technology is briefly
described in Section 2. Section 3 discusses challenges of integrating dynamic simulation
into the engineering workflow, illustrated by the GPU which is modelled with the
commercial tool UniSim. Section 4 presents approaches for the reduction and
integration of models implemented in different simulation tools, illustrated by the ASU
which is modelled with the in-house program OPTISIM.

2. Example: Carbon Capture and Storage (CCS)

CCS technologies allow separating carbon dioxide (CO ) when generating energy,
particularly in coal-fired power plants, and thus support the target of zero-emissions.
Linde Engineering is working closely with industry partners to develop technical
solutions for CCS [3], among them the oxyfuel process where coal is combusted in an
atmosphere of oxygen, produced by an Air Separation Unit (ASU), and CO . This means
that the nitrogen content in the resultant flue gas is significantly reduced, and the flue gas
primarily consists of CO and water. This CO -rich gas is further concentrated and
compressed in a Gas Processing Unit (GPU) and then transferred to storage. The ASU
and GPU are part of Linde Engineering's product portfolio [4,5].
2

3. Dynamic Simulation within the Plant Engineering Workflow

The complexity of new process steps and novel plant configurations calls for intensive
dynamic simulations right from project start. The basic task is to identify the most
appropriate process concepts for the ASU and GPU out of numerous process variants.
Both technical requirements such as guaranteed product purities at specified load change
rates, and business requirements such as lowest investment costs aligned with highest
energy efficiency and flexibility have to be satisfied. Originating from steady- state
design process models, dynamic simulation models are derived for verification of static
equipment design and development of control strategies. Some challenges arising from
integrating this approach into the plant engineering workflow are discussed using the
GPU design process as an example.
3.1. Communication between subject matter experts
Dynamic simulation can play a key role in a variety of engineering tasks - process
design, control and operational strategy design including startup and shutdown
procedures, safety and abnormal situation handling, etc. However, its use still faces
barriers as dynamic simulation is often considered an "expert tool" and is not easily
accessible to engineers of various backgrounds and design responsibilities. In order to
remove these barriers and facilitate efficient communication between subject matter
experts, Linde Engineering uses even at an early project stage Human Machine Interface
(HMI) visualization to develop graphical user interfaces for dynamic models. This
approach provides an intuitive and common environment in which process
configurations, control and operational strategies can be demonstrated and discussed. It
lets the user better focus on the relevant (e.g. operational) information than by using the
simulation software interface directly, which typically provides a detailed process design
view. For the GPU example, such a process visualization consisting of a process
overview (see Fig. 1) and a series of more detailed process area views has been
implemented in UniSim Operations, resulting in significantly improved acceptance of
the dynamic model.
In this particular GPU design [5] CO2 is separated from the oxyfuel flue gas stream using
a cryogenic approach - raw gas compression, CO 2 liquefaction via a main heat exchanger
and two separators close to the triple point of CO 2, vaporization and consequent product
compression. Before being partially used for adsorber regeneration, vent gas is routed
through the main exchanger and two expansion turbines, which drive booster
compressors in the raw gas stream and in one of the product streams. The process
objectives are to maintain a required CO 2 recovery rate while satisfying the product
purity specifications for a given range of feed conditions.

Figure 1. GPU Overview

3.2. Iterative nature of design procedure
It is generally accepted that the benefits of dynamic simulation tend to be largest if the
findings become available early enough in a project so that design changes can still be
incorporated. Yet, a high-fidelity dynamic model can only be developed if detailed design
information is available. To reconcile these opposing trends it is imperative to integrate
dynamic simulations tightly into the process design workflow. Initially, steady-state
design has not advanced sufficiently for specifying detailed simulation models for
equipment such as turbo machines, heat exchangers, or piping. Hence, the fidelity of the
initial dynamic models is low as e.g. constant efficiencies for turbo machines are
assumed. At this stage the focus of dynamic simulation is on evaluating general
operability and control concepts, and understanding the dominant dynamic modes of the
process. As the design advances, more equipment details get specified (e.g. performance
maps for turbo machines) which have to be transferred to the dynamic model. Therefore,
the simulation software needs to be flexible to allow for different modeling levels within
individual process units. Findings from dynamic analysis impact the static process
concepts, resulting in an iterative design procedure.
Typical applications of high-fidelity dynamic models are discussed below:
(i) Design evaluation: It is determined whether the design objectives are met during
transitions between steady-state design points, and how robust the design is with respect
to disturbances (such as varying feed and utility supply conditions) and equipment
failures (such as turbine trips or loss of cooling water). As an example, the GPU model is
subjected to a 30% rate reduction introduced over 7.5 min and a concurrent 2% reduction
in feed CO2 concentration. The simulation results in Figure 2 confirm that the product
purity specifications (>95% CO2, <1% O2) are maintained throughout the transition.
However, the specific energy consumption increases due to the lower efficiency of the
turbo machines which at around t=7min start to move into recycle mode.

Figure 2. GPU model predictions during a feed transition

(ii) Design improvements: Knowledge of the dynamic behavior can be used to improve
the design of the overall process as well as the individual units. As an example, the
temperature peak resulting from a switchover between adsorber beds impacts the
operation of the main heat exchanger. While the exchanger was initially designed
based on steady-state (worst-case) inlet temperature predictions, resulting in a certain
level of overdesign, the dynamic temperature predictions are used to further improve
the exchanger design.
(iii) Control strategy development: While the basic control strategies can be developed
and evaluated using a low-fidelity model, a thorough evaluation of control and
operational strategies requires both accurate process unit models and specification of
all control-relevant equipment and logic such as compressor anti-surge and expander
speed limiting controllers. For the GPU process, simulations covering a wide range of
operating conditions revealed the need for additional temperature control loops in the
vent gas path.

4. Model Reduction
Once dynamic simulation models have been developed for each of the oxyfuel process
units ASU, GPU and others, the next step is to combine theses models to a plant wide
dynamic model for operability analysis and control system design. The fact that each
industry partner contributing process units to the oxyfuel process usually applies its own
design and dynamic simulation software, calls for a strategy to efficiently combine
dynamic models from different dynamic simulation programs. Options to set up a plant
wide dynamic model composed of process unit models originating from commercial
simulation tools, e.g. UniSim and APROS, generic modeling software, e.g. gPROMS,
MATLAB and Dymola, or in-house developments, e.g. OPTISIM, are discussed.
Without identifying a specific target platform for integration in the current example its
selection must consider various aspects such as the effort of transferring dynamic models
to the target platform, reasonable computing times, acceptable level of model fidelity of
the transferred model, protection of each contributor's model know-how, etc. The latter
may be important in case the integrated model shall be available to all contributing
partners. What clearly needs to be avoided is the attempt to re-implement an existing
dynamic model in a new software tool.

Several options have been evaluated for transferring models from the original dynamic
simulation tool to a target platform for integration:
(i) Applying an operating point linearization of a nonlinear dynamic process unit model
in an equation based simulation tool, e.g. OPTISIM [2], results in a linear dynamic
state space model. The transfer to the target platform then includes a linear model
reduction step [6] to gain a reduced dimensional linear state space model.
(ii) For identification of a linear or nonlinear dynamic system, responses of the original
system to step inputs, PRBS (pseudo-random binary sequence), or other suitable
signals are generated [7]. This information is input to commonly available system
identification software for linear systems or e.g. artificial neural networks. Input
signals and system responses can be generated with models in any simulation tool.
(iii) A direct coupling of the original heterogeneous dynamic simulation tools in an
overall integration platform has been reported in [8].
For the oxyfuel example, the first two options have been pursued due to aspects of knowhow protection, reducing computing time and maintaining flexibility for integrating the
models into different simulation target platforms. Numerous reduction and identification
algorithms can be found in free and commercially available software tools. Either
approach to model transfer results in an encoded model of reduced order. Prior to
integration into the target platform, the range of validity of the reduced model must be
evaluated via comparison runs with the original models.

As an example, the transfer of the ASU model from OPTISIM to a linear reduced state
space model is explained. Figure 3 shows a typical ASU configuration for an oxyfuel
process [4]. The main AIR stream is compressed and cooled in the main heat exchanger
and fed to the bottom of the high pressure column. The product stream GOX with 95 %
oxygen is withdrawn from the bottom of the low pressure column and reheated in the
main heat exchanger. Nitrogen is removed as high pressure product PGAN or as low
pressure rest gas.

Figure 4. Load change simulation with nonlinear model and linear reduced model
The nonlinear model set up in OPTISIM consists of a set of 7482 differential and
algebraic equations (DAE). This model is linearized at the steady state operating point of
100% load to yield a linear DAE system. After eliminating the algebraic subsystem,
abalanced residualization algorithm for model reduction is applied to obtain a linear
ODE system of size 70. The original nonlinear model is compared to the linear reduced
model in various test runs. Figure 4 shows a typical system response in selected oxygen
concentrations of both models to changes in AIR and GOX flow rates.

5. Conclusion
The paper presented practical aspects of dynamic simulation from the perspective of a
plant engineering and contracting company. Typical challenges and selected solutions
were discussed and illustrated using the oxyfuel process as an example. It was shown
how dynamic simulation can improve the process and control design. Additional benefits
were realized by an early workflow integration and HMI visualization. Furthermore,
results of a successful model reduction were presented, which provides the basis for
integrating models from different simulation tools.

References
[1] G. Engl, A. Kroner, 2006, Success Factors for CAPE in the Engineering Practice of a Process
Plant Contractor, In: W. Marquardt, C. Pantelides (eds..): Computer-Aided Chemical
Engineering 21A, Proc. of ESCAPE-16 and PSE'2006, Elsevier, Amsterdam, 763-768.
[2] A. Kroner, 2006, An Engineering Company's Approach to Filling "CAPE Gaps" in Process
Simulation, In: W. Marquardt, C. Pantelides (eds.): Computer-Aided Chemical Engineering
21A, Proc. of ESCAPE-16 and PSE'2006, Elsevier, Amsterdam, 781-786.
[3] Linde AG, 2008, Carbon Capture and Storage, The Linde Corporate Responsibility Report
2008, 59-63.
[4] G. Beysel, 2009, Enhanced Cryogenic Air Separation: A Proven Process Applied to Oxyfuel,
In: S. Santos (ed.): Proc. of 1st Oxyfuel Combustion Conference, Cottbus, Germany.
[5] R. Ritter, A. Kutzschbach, T. Stoffregen, 2009, Energetic Evaluation of a CO2 Purification and
Compression Plant for the Oxyfuel Process, In: S. Santos (ed.): Proc. of 1 st Oxyfuel
Combustion Conference, Cottbus, Germany.
[6] A. C. Antoulas, D. C. Sorensen, and S. Gugercin, 2001, A survey of model reduction methods
for large scale systems, Contemporary Mathematics, AMS Publication, 280, 193-219.
[7] M. Pottmann, D. E. Seborg, 1992, Identification of nonlinear processes using reciprocal
multiquadric functions, Journal of Process Control, 2, 189-203.
[8] V. Kulikov, H. Briesen, R. Grosch, A Yang, L. v. Wedel, W. Marquardt, Modular dynamic
simulation of integrated process flowsheets by means of tool integration, Chemical
Engineering Science, 2005, 60(7), 2069-2083.