Newcastle University

Process Control 2

CME3008

Dr. Mark Willis

Assignment 1

24/11/2016

1 | Page
Introduction
A bioreactor with a dilution rate between 0<D(t)<1.5 is given as an open loop unstable system.
The aim of this assignment is to create a controller using the methods learned in class for PI(D)
controllers to maintain the biomass concentration at its theoretical normal operating value of 2.55
[gl-1]. The objective is to preserve the bioreactor in a stable non-oscillatory form, responding to
our desired output with the controllers created. The system should have the smallest time delay
possible and the substrate concentration will be considered as it should not fluctuate to a point
where the costs of operation in the bioreactor may rise.
Once the controller has been created with the appropriate parameters, it will need to be tested
against the different set point changes to understand its practically in the real life to maintain
biomass at its optimal concentration.
Process Structure
The system identification toolbox in Simulink has been used to estimate the process model,
Figure 1. It also represents the adopted procedure for input and output data collection. The
output was represented in terms of deviation by subtracting the initial steady state 2.55 g/L.

Figure1 Simulink simulation for data collection and utilization

Random step changes of 0.9 with 0.01 variance were made as shown in figure 2 and 3, the model
and the process responses are plotted in deviation form.

Figure2 Estimated Process Model Figure3 Random step changes in Dilution

2 | Page
The process was estimated to be a First Order Process Dead Time (FOPDT) equation (1), as it
has a time delay of 5 seconds in the system’s response. The transfer function’s gain was negative
as expected since the response of the system was opposite to the input change.

(1) (2)
In addition, the process was modelled as a Second Order Plus Dead Time (SOPDT) equation (2)
transfer function using a 0.9 step change. The damping factor is approximately equal to 1 and
that means the model is critically damped. SOPDT fits within the range as shown:
FOPDT is more feasible to estimate and can be applied to design a PI or PID controller, the
model will not oscillate. Second order transfer function is expected to have a better response;
however, it is restricted for only PID controller design, the model may oscillate depending on the
damping factor. Since the process is nonlinear, there will be a finite Plant Model Mismatch
(PMM). The PMM will be tolerated, assuming the system will be operating within the steady
state and the output will not deviate significantly.
For this particular bioreactor the FOPDT model has been chosen because it is more viable to
work with a system that does not oscillate and responds well to any controller such as PI(D) and
is not restricted to PID as in SOPDT.
The accuracy of the model is proven by taking the average of the different step changes and
identifying the output of the model stays the same as the output of the actual model. Thus, the
model behaves as expected.
The limitations of the model are based on having a nonlinear process and using a FOPDT which
is used for linear systems. And since the process is non-linear, there will be a finite plant model
mismatch. The operation of the system is limited to a dilution of 0.8 to 1.
Control Law Development Using Direct Synthesis Tuning Procedure

Through the direct synthesis method is possible to identify a strategy for the controller and its
parameters in order for the system to stay within the required form. Thus giving the expected
behavior.

FOPDT

Using Taylor Series is a good approach to identify the

control laws of the system because it allows to find
the PI of the FOPDT when approximating
−θps
e ≈ 1−θ p s.
Using the direct synthesis for a FOPDT using Taylor
approximation:

Figure 4 Process Response for a PI Controller

3 | Page
τI=0.001736

τp
Our closed loop time constant is λ= , because time constant of the controller needs to be twice
2
as small as time constant of the process, in order to have an appropriate controller.
First Order Plus Dead Time
Another approach to create our controller is
through the use of Direct Synthesis with Padé
approximation, to obtain the PID controller.
Using a FOPDT Padé approximation
controller strategy assuming time delay is less
than the controller settling time and less than
process time constant (θp<<τc & θp<<τp),
therefore it is possible to neglect the quadratic
term in the denominator, resulting in:
τ p s+1
Gc =

( )
1
1−
2 θp s
K p∗ τ c +1−
1
1+
2 θp s Figure 5 Process Response for a PID Controller
1
(τ ¿¿ p s+ 1)(1+ θ p s)
2
Gc = ¿
Kp¿¿
1 1 2
(τ ¿ ¿ p+ θ p) s +1+ τ p θ p s
2 2
Gc = ¿
( 1
2
1
2
2
K p (τ ¿¿ c + θ p ) s+ 1+ τ c θ p s ¿−(1− θ p s)
1
2 )
1 1 2
(τ ¿ ¿ p + θ p ) s+1+ τ p θ p s
2 2
Gc = ¿
Kp¿¿
1 1 2
(τ ¿ ¿ p + θ p ) s+1+ τ p θ p s
2 2
Gc = ¿
K p ( τ c +θ p ) s
1

[ (( ]
(τ ¿ ¿ p + θ p )
2 1 τ θ
Gc = 1+ + p p s ¿

( ) ))
K p ( τ c +θ p ) 1 2 τ p +θ p
τ p+ θp s
2
Smith & Corropio: (3)

4 | Page
For our model
K c =-33.8806 τ I =3.3491 τ D =2.5

SOPDT

For the step by step approach in obtaining the Controller gain (Kc), Integral time (τI) and
Derivative time (τd) constants using the direct synthesis method for the SOPDT, applying Taylor
Series approximation.
(4)

K c =44.148 τ I =3.976 τ D =0.994

For FOPDT is better to use direct synthesis with Padé approximation as the results are more
accurate and lead to a better output than direct synthesis with Taylor series approximation. Padé
approximation for FOPDT outputs similar results as Taylor Series approximation for SOPDT,
therefore we can conclude our PID controller for FOPDT is a very accurate controller. The ratio
of the time delay to the time constant is larger than 1, which means the process system has a
large dead time (Liptak, 2006). This is seen by the significant undershoot of the system response
in figures (4 and 5), where the new steady state is reached after 25 seconds. Therefore, it is
confirmed that there is a need of a tuning method to compensate the large time delay of the
controller.
Smith Predictor
Smith predictor is a systematic controller that is
commonly used for systems with a long dead time.
The controller performance is crucially affected by
the long-time delay. This is because the controller
will not tolerate the time delay as the output of the
system doesn’t match the set point. For instance,
when opening a valve of a water line, it will take a
longer time for the water to reach the outlet in a very
long line. The time delay in the water line may
enforce the operator to open up the valve even more,
“for being impatient”, resulting in a high flow of
water as it reaches the end of the line. This is
Figure 6 Process Response for a PI Controller with
analogous to PI(D) controller’s response for a system
Smith Predictor
with a large dead time.
Smith predictor controller approached by making an
inner loop for the process model in the conventional
feedback loop of the process (VanDoran, 2015). The
output of the inner loop is subtracted from the output
of the process, where it results in an estimated
disturbance on the system. The measured disturbance
5 | Page
is added to the output of the process model excluding time delay, and then fed back to the set
point as shown in figure Figures 6 and 7. Mathematically, Smith predictor generates process
model output with disturbance and applies it to the setpoint for error computation.
Integral Square Error of the PI controller is 122 and
Figure 7 Process Response for a PID Controller
PID controller 84.77 by implementing the Smith with Smith Predictor
Predictor the ISE dropped to 6.6 for the PI controller
and to 6.5 for the PID of FOPDT.

Conclusions
FOPDT is the model by choice as it allows us to create controllers that would maintain the
biomass concentration in its optimal rate without moving the dilution rate out of its range of
process. The preferred controller method used was PID with Direct Synthesis through Padé
approximation and Smith Predictor as this allows to minimize the large dead time of the
controller and gives the desired performance for the output of the process model.
SOPDT is a more accurate model however it is more likely to create oscillation and only allows
PID controllers, which are more expensive in the industry than PI controllers.

References

[1] VanDoren, V. (2015, February 17). Overcoming process deadtime with a Smith Predictor.
Retrieved November 03, 2016, from http://www.controleng.com/single-article/overcoming-
process-deadtime-with-a-smith-predictor/8c727a1371eb45011801350175606812.html

[2] Luyben, M. L., & Luyben, W. L. (1997). Essentials of process control. New York: McGraw-
Hill.

[3] Liptak, B. G. (2006). (3rd ed., Vol. Three). New York, NY: CRC Press.

6 | Page