1267

A publication of

CHEMICAL ENGINEERING TRANSACTIONS

The Italian Association
VOL. 32, 2013 of Chemical Engineering
Online at: www.aidic.it/cet
Chief Editors: Sauro Pierucci, Ji J. Kleme
Copyright 2013, AIDIC Servizi S.r.l.,
ISBN 978-88-95608-23-5; ISSN 1974-9791

Systematic Modelling of Flow and Pressure Distribution

Heinz A Preisig
Department of Chemical Engineering, Norwegian University of Science and Technology, Trondheim, Norway

[email protected]

In any plant, it is the pressure that drives convective flow of mass from one part to another. A systematic
method of modelling the flow and the pressure distribution is derived based on a finite-volume approached
captured in a graph. This approach is conform with our network modelling method visual modelling.

1. Background
Most process models are based on a network of finite volumes. Chemical engineers refer to it as the control-
volume approach (Preisig, 2010). We look at the network as a graph, where the volumes represent the
capacities and the arcs the transfer of extensive quantities, more precisely mass flow, usually a vector of
component mass flows, energy flow, usually conductive and radiation heat flow, mechanical work, volume
work and the like. Whilst mass has a considerable inertia, the pressure distribution is happening at the speed
of sound, thus comparatively very fast. In most cases, the dynamics of the pressure distribution is not of
interest in chemical processes and one assumes it to just happen, technically speaking one assumes event-
dynamics. The exception is the description of explosions, detonations any system that operates on the time
scale of pressure wave propagation in a system. But why do we need the pressure distribution? First reason is
that convective flow is driven by the pressure difference and the second one is that the material properties are
a function of pressure as it enters the equation of state and thus the energy functions.

2. A Sample System
To illustrate the approach, we introduce a simple, but representative common part of a plant (Fig. 1). It is a
near barometric pressure inside. In addition the tank has a feed pipe connected to the water supply on the top,
and the bottom outlet is connected to the drain at barometric pressure.
The mechanics of the process are apparent: As one puts water into the tank it starts filling up, whilst the water
outflow is driven by the pressure in the tank. Latter is essentially the barometric
pressure plus the hydrostatic pressure exerted by the water. The flow into the tank is
a free jet that extends from the end of the pipe to the water surface. The purpose of
the breathing pipe is to allow for air to go in and out of the tank so as to equalize the
inside pressure with the outside pressure. This is a dynamic process and if the tank
level changes fast, the breathing pipe's resistant will have the effect that the pressure
inside and outside are not equal.

The modelling of the process is done step by step with the first one being an
abstraction of the plant as a network of capacities and flows of extensive quantities.
2.1 A first topology
The modelling approach we call visual modelling. The plant model is represented
as a graph choosing the nodes to represent capacities and the arcs to represent
flows of extensive quantities. When establishing this graph, we make the first set of
time-scale assumptions.
Any system we describe uses three time domains, one in which the associated Figure 1:
capacities are considered constant, one in which they are dynamic and change Sample process
1268

visibly with time, and one in which things just happen, being event-
dynamic. The constant systems, we call reservoirs. Shown as half
circles they are open on one side extending to infinity. For
dynamic systems we have two cases to consider, namely
distributed systems and lumped systems. Lumped systems are
characterised by intensive properties not to be a function of the
position in contrast to the distributed systems where the intensive
properties are a function of the position. Both type of systems are
of finite dynamic nature and are mathematically formulated as
differential equations. Lumped systems are ordinary and
distributed systems are represented by partial differential
equations. In the third domain we have event-dynamic systems,
which, when lumped are shown as simple bars and when
distributed as rectangles. The initial topology always contains
maximal information. Any change towards more detail requires an
extension of the graph splitting capacities and adding more
transfers, make the granularity of the model finer (Preisig, 2010).
Our first and maximal graph is shown in Fig. 2. Here the oval
nodes represent the distributed capacities as labelled, whilst the
bars represent the boundaries between the respective capacities.
2.2 More time-scale assumption
Further simplifications are based on additional time-scale
assumptions. In this case, the next step requires a split of the Figure 2: A first graph
plant model into two separate time scales. Since we assume that the pressure wave propagate very fast, we
get an event-dynamic model for the pressure distribution, whilst the dynamics of the mass distribution
happens in the visible dynamic range. Both are embedded in a set of reservoirs that represent the process-
embedding environment, which are considered constant in terms of their intensive properties and infinitely
large in terms of their capacity. So we first split the description into a model for the event-dynamic part and the
dynamic part, the first one being the momentum propagation and the second the mass.
The left part of Fig. 3 shows the event-dynamic model, the right the dynamic model. As mentioned, we use
rectangles for the distributed capacities and circles for the lumped system. The black rectangles show
systems that reduce to a plug-flow only and are reduced to a simple dead-time for the transport; whilst the
coloured rectangles also consider the distribution in the intensities, in particular the pressure. The time-scale
assumptions can be extended as demonstrated on the water jet, where on the right-hand-side it is assumed
instantaneous so no significant dead-time associated with the water passing from the end of the pipe to the
surface, whilst on the left-hand-side, it is modelled as an event-dynamic distributed system. We could consider
making the same assumption for the pipes, thus the water pipe in and out and the breathing pipe. This then
results in what most people would probably write as their very first model for the behaviour of the plant in
terms of mass, namely, water accumulated per unit time in the tank is the difference between what is coming
in and what is going out.

3. Model Equations
The mathematical model is assembled from the models of the capacities and the model of the transfers. The
equations for the different parts are collected in the table below. The table's first column contains a description
of the model component, the second introduces a label for each equation which we use below to discuss the
overall equation pattern. The third contains the equation.
The first equation describes the mass conservation in vector form for a single homogeneous material for a
lumped system. It simply says that the accumulation is equivalent to the sum of all the physical inflows minus
the physical outflows. The representation makes use of the underlying directed graph for the physical
topology. There we introduce the reference co-ordinates for each flow, represented by the directed arcs, the
arrows. The matrix is the row of the incidence matrix of the graph for system s. The incidence matrix has
as row the nodes (systems / capacities) and in the columns the arcs (transports). So for the mass balance of
the tank the mass conservation equation is .
 1269

Figure 3: Splitting the time-scales: left event-dynamic, right dynamic

Note on how we use the indexing scheme to indicate the flows and its direction: The source is the first part of
the index using the identifier of the system being at the tail of the directed arc. The vertical bar is the separator
and the second index is the identifier of the sink, the head of the directed arc. The row for the fluid in the tank
of the incidence matrix thus is all zeros except the columns for the flow 3|W and W|4.
For the pressure distribution we use the event-dynamic model for which the slow model is entering as a snap
shot, that is, an implicit singular perturbation is done assuming a pseudo steady state for the tanks mass.
The geometry relates the volume to the level in the tank and the material properties related the state mass to
the volume, which is one of the critical pieces of information required.
Table 1Model equations

Component description Symbol Equation

Mass of lumped system

Mass balance lumped (dynamic capacity s)

Mass balance lumped / distributed event-dynamic
Energy balance (dynamic capacity s)
Energy balance (event-dynamic s)
Mass flow
Kinetic energy flow m

Potential energy flow m

1270

Component description Symbol Equation

Volume work flow m

Friction work over transport system s

Pressure drop due to friction transport system s

for cylindrical geometry and incompressible fluid
Darcy-Weisbach model
Friction factor model for laminar flow

Friction factor model for turbulent flow

Friction factor model for highly turbulent flow

Reynold number

Geometrical relation

Geometrical relation

Geometrical relation

Geometrical relation

Geometrical relation

Property
Rigid container

Note that we distinguish between accumulation and

flow. For the first we use the dot-decorator, whilst for
the second the hat-decorator. Obviously both have the
same physical units.

The plant model consists of the components:

The fluid being incompressible for all practical purposes.
Pipes that are assumed to be cylindrical.
Equations

Assumptions (A)
Facts (F) cylindrical:

Tank
Equations
Constant volume
Tank event-dynamic liquid phase or gas phase:
Equations

Assumptions (A)
Facts (F) cylindrical: no friction:

Ditto, but dynamic liquid phase:

 1271

Equations
Assumptions (A)
Facts (F) cylindrical: no friction:

Ditto, but dynamic gas phase:

Equations
Assumptions (A)
Facts (F) Negligible: cylindrical: no friction:

negligible:

Free jet that is assumed to be cylindrical

Equations
Assumptions (A)
Facts (F) Negligible: cylindrical: no friction:

negligible:

3.1 Assembling the model for the plant

In view of the page limits we continue with an event smaller, simplified version of
the given plant: It consists of a series of flow units that transport an incompressible
fluid. The model assembles from the mass and energy balance for the tubes and
the boundaries between the tubes:
The plant's position
Pipes A,B,C in the state space
is defined by the
Interfaces 2,3 conditions on the
boundaries, which
Interfaces 2,3
in this case are the
pressures at the
interfaces 1 and 4. These are the pressures of the respective connected reservoirs.
In more complex models, also the other intensive properties of the reservoirs enter
the definition space, for example the composition. Further, we require the geometry
of the pipe system, here we assumed cylindrical pipes. In addition we need the
location of the interfaces in the space, thus the relative height measurements.
The mass balances for the interfaces equal the flows, thus there is only one overall
flow in the system. The energy balances of the interfaces equal the pressure on
either side of the interfaces. This leaves 3 quantities to be computed, namely the
mass flow through the system and the pressures at the interfaces 2 and 3. These
can be obtained from the three energy balances for the pipes.
For laminar flow systems, the friction term is linear in the mass flow. So one can
rewrite the equations in a matrix format:

with the vector being a vector of constant friction coefficients.

Figure 4: Pipes only
The open tank adds the element of branching in the network and one of the nodes
being dynamic with respect to mass, namely the contents.

Capacities

Capacity W
1272

Capacity A

Capacity A & W

Properties
Interfaces

Interfaces

The main feature here is that the level in the tank is a function of the slow part, namely the dynamic mass
balance. Also the flow through the breathing pipe is determined by the dynamics in the tank. The dynamics of
the water is given by the flows in and out, whilst the dynamics of the gas phase, the air, is driven by the
change in the volume of the water in the tank and the fact that the total volume is constant. The capacity
model applicable to P, J, O, P uses the mass balance for the same systems and the density.

4. Conclusions
The analysis of flow systems requires first a split into two separate descriptions for mass and mechanical
energy. Whilst the mass balances are dynamic, the mechanical energy balances are event-dynamic assuming
that the pressure propagates much faster than the level changes. Latter is used by implementing a singular
perturbation assumption for all the systems or what in chemical engineering is usually referred to as pseudo-
steady state.
The presented analysis can be further abstracted, which exploits the structural properties further, though it
gets harder to communicate. Specifically, the depicted graph can be seen as a bipartite graph, with one set of
nodes being the interfaces and the other being the capacities. The analysis shows that the fields appear as
the driving forces for the transport in one half of the bipartite graph. The analysis though takes more space
than we have available here, so we must leave this extension to a later exposition.
The fact that one requires the boundary conditions of the event-dynamic system to be defined seems quite
obvious even though that is often a piece of intense discussion in practice.
How can these results be used? One of the main applications is in alarm and warning handling. Large plants
suffer of alarm flooding and modern control systems are starting to do active alarm hiding by constructing
alarm management systems (Hollender et al, 2007). Latter found some significant industrial interest for a good
number of years, but has manifest itself clearly with the two patents (Skld et al, 2010) and (Thurau, 2011).
The apparent other application is Hazop analysis as documented in (Venkatasubramanian, 2000) and
(Dunja, 2010).

References
Hollender, M. and Beuthel, C.; 2007, "Intelligent alarming," ABB Review (:1), 2007.
Skld, M and Hansen, M and Gjerseth, J 2010, US Patent 7,692,537,Method and system for automatically
deciding what alarm, generated in an industrial plant, to hide or to present to an operator,
Thurau, O - US Patent 7,953,503, Method for operating a process plant, process plant and computer program
product, 2011
Preisig, Heinz A , 2012 HAZOP an automaton-inspired approach, Proceedings of the 22nd European
Symposium on Computer Aided Process Engineering, 17-20, June 2012, London, 1242-1246.
Venkatasubramanian, Venkat, 2000, Intelligent systems for HAZOP analysis of complex process plants; Comp
Chem Eng 24, 22912302.
Preisig, H.A.; 2010, Constructing and maintaining proper process models, Comp & Chem Eng Vol. 34(9), pp.
1543-1555.
Dunja, Jordi et al., 2010, Hazard and operability (HAZOP) analysis. A literature review; Journal of Hazardous
Materials 173 (2010) 1932.