ILLUSTRATIVE EXAMPLE: A BLENDING PROCESS

A simple blending process is used to introduce some important issues in control system design. Blending
operations are commonly used in many industries to ensure that final products meet customer specifications. A
continuous, stirred-tank blending system is shown in Figure 1. The control objective is to blend the two inlet
streams to produce an outlet stream that has the desired composition.
Stream 1 is a mixture of two chemical species, A and B. We assume that its mass flow rate w1 is
constant, but the mass fraction of A, , varies with time. Stream 2 consists of pure A and thus = 1. The
mass flow rate of Stream 2, , can be manipulated using a control valve. The mass fraction of A in the exit
stream is denoted by x and the desired value (set point) by .
Thus for this control problem, the controlled variable is x, the manipulated variable is , and the
disturbance variable is .
Next we consider two questions:
Design Question: If the nominal value of is , what nominal flow rate is required to produce the
desired outlet concentration, ?
To answer this question, we consider the steady-state material balances:
Overall balance:

Component A balance:

The over-bar over a symbol denotes its nominal steady-state value, for example, the value used in the
process design.
According to the process description, and ·
Solving Eqn. (1) for w, substituting these values into Eqn. (2), and rearranging gives:

Eqn. (3) is the design equation for the blending system. If our assumptions are correct and if ,
then this value of will produce the desired result, . But what happens if conditions change?
Control Question: Suppose that inlet concentration varies with time. How can we ensure that the
outlet composition x remains at or near its desired value, ?
As an example, assume that
increases to a constant value that is larger
than its nominal value, .
It is clear that the outlet composition
will also increase due to the increase in inlet
composition. Consequently, at this new
steady state, x > .
Next consider some strategies for
reducing the effects of on x.

Figure 1: Stirred-tank blending system

Method 1: Measure x and adjust
It is reasonable to measure controlled variable x and then adjust accordingly.
For example, if x is too high, should be reduced; if x is too low, should be increased. This control
strategy could be implemented by a person (manual control). However, it would normally be more convenient
and economical to automate this simple task (automatic control).
Method 1 can be implemented as a simple control algorithm (or control law),

where is a constant called the controller gain, and and , indicate that and change with time.
Eqn. (4) is an example of proportional control, because the change in the flow rate, , is
proportional to the deviation from the set point, . Consequently, a large deviation from set point
produces a large corrective action, while a small deviation results in a small corrective action.
Note that we require to be positive because must increase when decreases, and vice versa.
However, in other control applications, negative values of are appropriate.
A schematic diagram of Method 1 is shown in Figure 2. The outlet concentration is measured and
transmitted to the controller as an electrical signal. (Electrical signals are shown as dashed lines in Figure 2.)
The controller executes the control law and sends the calculated value of to the control valve as an electrical
signal. The control valve opens or closes accordingly.

Figure 2: Blending system and Control Method 1 Figure 3: Blending system and Control Method 2

Method 2: Measure , adjust

As an alternative to Method 1, we could measure disturbance variable and adjust accordingly.
Thus, if > , we would decrease so that < If < , we would increase . A control law based
on Method 2 can be derived from Eqn. (3) by replacing with and with :

The schematic diagram for Method 2 is shown in Figure 3. Because Eqn. (3) is valid only for steady-
state conditions, it is not clear just how effective Method 2 will be during the transient conditions that occur
after a disturbance.
Method 3: Measure and , adjust
This approach is a combination of Methods 1 and 2.
Method 4: Use a larger tank
If a larger tank is used, fluctuations in will tend to be damped out as a result of the larger volume of
liquid. However, increasing tank size is an expensive solution due to the increased capital cost.
Dynamic Model:
We have developed a steady-state model for a stirred-tank blending system based on mass and
component balances. Now we develop an unsteady-state model that will allow us to analyze the more general
situation where process variables vary with time. Dynamic models differ from steady-state models because they
contain additional accumulation terms.
As an illustrative example, we consider the isothermal
stirred-tank blending system in Fig. 4. It is a more general
version of the blending system in Fig. 1 because the overflow
line has been omitted and inlet stream 2 is not necessarily pure A
(that is, ≠ 1). Now the volume of liquid in the tank V can vary
with time, and the exit flow rate is not necessarily equal to the
sum of the inlet flow rates. An unsteady-state mass balance for
the blending system in Fig. 4 has the form

Figure 4: Stirred-tank blending process

The mass of liquid in the tank can be expressed as the product of the liquid volume V and the density ρ.
Consequently, the rate of mass accumulation is simply d(Vρ)/dt, and one can write

where and are mass flow rates.

The unsteady-state material balance for component A can be derived in an analogous manner. We
assume that the blending tank is perfectly mixed. This assumption has two important implications:
(i) there are no concentration gradients in the tank contents and
(ii) the composition of the exit stream is equal to the tank composition.
The perfect mixing assumption is valid for low-viscosity liquids that receive an adequate degree of agitation. In
contrast, the assumption is less likely to be valid for high-viscosity liquids such as polymers or molten metals.
For the perfect mixing assumption, the rate of accumulation of component A is d(V x)/dt, where is the
mass fraction of A. The unsteady-state component balance is

Equations (6) and (7) provide an unsteady-state model for the blending system.
The corresponding steady-state model was derived as.

It also can be obtained by setting the accumulation terms in Eqs. (6) and (7) equal to zero.
Note: The nominal steady-state conditions are denoted by and , and so on. In general, a steady-state model
is a special case of an unsteady-state model that can be derived by setting accumulation terms equal to zero.
A dynamic model can be used to characterize the transient behavior of a process for a wide variety of
conditions. For example, some relevant concerns for the blending process: How would the exit composition
change after a sudden increase in an inlet flow rate or after a gradual decrease in an inlet composition? Would
these transient responses be very different if the volume of liquid in the tank is quite small, or quite large, when
an inlet change begins? These questions can be answered by solving the ordinary differential equations in (6)
and (7) for specific initial conditions and for particular changes in inlet flow rates or compositions.

Next, we show that the dynamic model of the blending process in Eqs. 2-2 and 2-3 can be simplified and
expressed in a more appropriate form for computer simulation. For this analysis, we introduce the additional
assumption that the density of the liquid, p, is a constant. This assumption is reasonable because often the
density has only a weak dependence on composition. For constant , Eqs. (6) and (7) become

Equation (9) can be simplified by expanding the accumulation term using the "chain rule" for
differentiation of a product:

Substitution of the mass balance in (8) for in (10) gives

After canceling common terms and rearranging (8) and (11), a more convenient model form is obtained:

The dynamic model in Eqs. (12) and (13) is quite general and is based on only two assumptions: perfect
mixing and constant density. For special situations, the liquid volume is constant (that is, dV/dt = 0), and the
exit flow rate equals the sum of the inlet flow rates, . For example, these conditions occur when
1. An overflow line is used in the tank as shown in Fig. 1.
2. The tank is closed and filled to capacity.
3. A liquid-level controller keeps V essentially constant by adjusting a flow rate.
In all three cases, Eq. (12) reduces to the same form as Eq. (1), not because each flow rate is constant, but
because at all times.
The dynamic model in Eqs. (12) and (13) is in a convenient form for subsequent investigation based on
analytical or numerical techniques. In order to obtain a solution to the ODE model, we must specify the inlet
compositions ( and ) and the flow rates ( , and ) as functions of time. After specifying initial
conditions for the dependent variables, V(0) and x(0), we can determine the transient responses, V(t) and x(t).
TRANSFER FUNCTION MODELS:
For simplicity, we make the following assumptions:
1. Liquid density and volume V are constant.
2. Flow rates , and are constant.
Then the component balance in Eq. (7) becomes

Case I: Inlet concentration varies while is constant

We will derive a transfer function model between exit composition x and inlet composition , starting
with Eq. (14). The steady-state version of (14) is

where the bar over a symbol denotes a nominal steady-state value. Subtracting (15) from (14) gives

where the two deviation variables (sometimes called perturbation variables) are defined as

Because is a constant, it follows that

Substituting (18) into (16) gives the solute component balance in deviation variable form:

Assume that the blending system is initially at the nominal steady state. Thus, and .
Taking the Laplace transform of Eq. (19) gives

where, and .
Rearranging gives the transfer function G(s) between the exit and inlet and compositions:

It is useful to place the transfer function in a standard form by dividing both the numerator and the
denominator by :

where constants and are defined as

Case II: Both inlet concentrations, and , vary
For the case of two input variables, and , two transfer functions are needed to describe their effects
on output variable . Their derivation is analogous to the derivation for Case I.
For this case, the steady-state version of (14) can be written as

Subtracting (24) from (14) and introducing deviation variables gives

Again assuming that the blending system is initially at the nominal steady state, taking the Laplace
transform of Eq. (25) gives

which can be rearranged as

where is defined as .
In order to derive the transfer function between and , assume that , is constant at its
nominal steady-state value, Therefore, , , and (27) reduces to the previous transfer
function relating and (see Eq. (22)).

Similarly, the transfer function between and can be derived from (26) and the assumption that is
constant at its nominal steady-state value, .

The models in (28) and (29) are referred to as first order transfer functions, because the denominators
are first-order in the Laplace variables. Three important aspects of these derivations are:
1. A comparison of (27) to (29) shows that the effects of the individual input variables on the
output variable are additive. This result is a consequence of the Principle of Superposition for linear models.
2. The assumption of an input being constant in the derivations of Eqs. (28) and (29) seems
restrictive but actually is not, for the following reason. Because a transfer function concerns the effect of a
single input on an output, it is not restrictive to assume that the other independent inputs are constant for
purposes of the derivation. Simultaneous changes in both inputs can be analyzed, as indicated by Eq. (27).
3. A transfer function model allows the output response to be calculated for a specified input
change. For example, Eq. (28) can be rearranged as