BACHELOR OF ENGINEERING (HONS.
) CHEMICAL
CEEH2205B
CHEMICAL PROCESS CONTROL
(CPE 501)
MINI PROJECT II
GROUP 6
Prepared by:
NAME STUDENT ID
AFIQAH AZZAHRA BINTI ZULKAFLY 2022664028
FATIMA ZAHRA BINTI ZAKI 2022605144
NOR ALIA AMIRA BINTI ZAILAN 2022616436
Prepared for:
SIR MUHAMAD FITRI OTHMAN
Chemical Process Control Lecturer
Date of Submission: 24th January 2025
A. SYSTEM IDENTIFICATION- NOR ALIA AMIRA BINTI ZAILAN
System identification is an essential process in control system engineering. It involves creating
mathematical models to represent how dynamic systems behave, based on actual data collected from the
system. These models help us understand the relationship between the input, which is the variable we
control, and the output, which is the response we want to manage. Instead of relying solely on theoretical
equations, system identification uses real-world data to capture how systems function, making it
especially useful for complex or less well-understood processes.
Dynamic systems are unique because their outputs depend on both current and past inputs, which
adds a layer of complexity to modeling them. The models used in system identification can either be
linear, like transfer functions, or nonlinear, which are better suited for more complicated behaviors. Linear
models are simpler and easier to work with, making them a common choice unless the system demands a
more advanced approach.
One of the most important aspects of system identification is finding the right balance between
accuracy and simplicity. Accuracy ensures the model closely matches the observed data, which is crucial
for making reliable predictions and controlling the system effectively. However, overly complex models
can overfit the data, meaning they perform well on the data they were trained on but struggle to generalize
to new scenarios. Simpler models, on the other hand, are easier to understand, quicker to compute, and
more robust to noise, making them ideal for real-time applications.
The process of system identification typically starts with collecting high-quality input-output data
from the system. For instance, if you're controlling the temperature in a preheater, the input might be the
flowrate of the heating medium (e.g., in m³/sec), and the output could be the temperature in degrees
Celsius. After collecting data, the next step is selecting a suitable model type. Common options include
linear models like ARX (Auto-Regressive with eXogenous inputs) or transfer functions, which are
straightforward and effective for many systems. Once the model type is chosen, algorithms are used to
adjust its parameters so the model output matches the real-world data. Finally, the model is tested on new
data to ensure it performs well and is robust.
For example, in a preheater temperature control system, you might use the flowrate as the input
and the temperature as the output. By applying system identification techniques, you can derive a transfer
function that accurately describes how changes in flowrate affect the temperature. Striking the right
balance between accuracy and simplicity allows engineers to create models that are both effective and
practical, ensuring smooth system operation (Introduction to System Identification, n.d.).
In summary, system identification is a powerful tool that helps us model, predict, and control
dynamic systems using real-world data. By focusing on both accuracy and simplicity, it ensures that these
models are not only reliable but also easy to use, enabling engineers to optimize and automate processes
across a wide range of industries.
Method:
1. Import input (flowrate) and output (temperature) data sheet into MATLAB.
2. Use the System Identification Toolbox function to create a model using the command
“ident”.
3. Plot trend for input (u1) and output (y1) data.
4. Estimate first, second and higher-order transfer function.
5. Select the most fit to estimation transfer function for the time-series dynamic data.
Results and Discussion:
Figure 1- Input (u1) and Output (y1) trend.
A transfer function developed using MATLAB's System Identification Toolbox serves as the process
model.
Order Transfer Function Fit to estimation data, % FPE (error), %
1st 1587 -2297 1385.55
𝑠 + 0.03725
2nd 2
59410 -4106 4469.62
𝑠 + 3.949𝑠 + 1.491
Highest 2
42050𝑠 + 552 -514.6 97.7215
𝑠 + 1.059𝑠 + 0.01889
Table 1- Optimum Transfer Function
Table 1 presents the best transfer functions generated for first, second, and higher-order models. Based on
the data, the higher-order transfer function achieved the highest percentage fit to the estimation data at
-514.6%, compared to -4106% for the second-order model and -2297% for the first-order model. With
such a high fit percentage, the higher-order model appears to be nearly ideal for estimating data values.
However, accuracy isn't the only factor to consider when choosing a reliable transfer function; error
values also play a significant role.
Table 1 also compares the Final Prediction Error (FPE) of the models. The higher-order transfer function
has the lowest FPE, followed by the first-order models and second-order. While the higher-order model
outperforms the others in terms of both fit percentage and error, its suitability must also be evaluated
based on the overall behavior of the system, including the shape of the response graph.
In conclusion, while the higher-order transfer function provides the best fit and lowest error, it strikes a
better balance between simplicity and accuracy for simulating the feedback control system in the
temperature control process. Therefore, the higher-order transfer function is the most appropriate choice
for modeling the temperature control time-series data.
B. CONTROLLER DESIGN AND TUNING - FATIMA ZAHRA BINTI ZAKI
(i) Block diagram to produce the process output response change with time when a unit step change is
made in the process input.
Figure 2- Block diagram
(ii) Process output response change with time — unit step change is made in the process input
Figure 3- Output response change with time
The system response, as depicted in the plot of Process Variable (PV) vs. Time, can be analyzed using key
dynamic parameters: rise time, overshoot, undershoot, and settling time. These parameters provide critical
insights into the system’s performance and how effectively it responds to changes in its setpoint. Below is
the justification of each parameter based on the observed system behavior.
1. Rise Time:
Rise time refers to the time taken for the Process Variable (PV) to move from 10% to 90% of its final
value for the first time. In this system, the PV initially decreases rapidly, indicating the system's quick
response to the control action. After this sharp initial decrease, the time it takes for the system to approach
its steady-state value defines the rise time. This is crucial as it indicates how quickly the system can react
and settle to the desired state after an initial disturbance.
2. Overshoot:
Overshoot occurs when the PV exceeds the final steady-state value during its transition to stability. In this
case, the plot shows that after the initial decrease, the PV rises slightly above the steady-state value before
beginning to stabilize. The maximum value reached above the steady-state before the system settles is
considered the overshoot. This is a common characteristic of underdamped systems, where the control
system overcompensates initially before stabilizing. The presence of overshoot suggests that while the
system is responsive, it may be prone to overreaction during transient conditions.
3. Undershoot:
Undershoot refers to the scenario where the PV temporarily falls below its final value before rising to
settle at the steady-state value. In this system, there is no significant undershoot visible in the plot. The
PV initially drops, but it does not dip below the final value before stabilizing. The absence of an
undershoot indicates that the system does not significantly undercompensate after a disturbance, which is
beneficial for maintaining control without unnecessary fluctuations.
4. Settling Time:
Settling time is the time it takes for the PV to remain within a specified range (typically 2% or 5%) of the
final value and stay there for the remainder of the observation period. In this system, after the initial
fluctuations and overshoot, the PV stabilizes around its final value and remains within a narrow range.
This steady behavior marks the settling time, which is important for understanding how long it takes for
the system to reach and stay within the desired operational state. A shorter settling time indicates faster
stabilization, which is desirable for ensuring the system does not take excessive time to adjust.
Conclusion:
Based on the observed system response, the rise time, overshoot, and settling time all reflect the system's
dynamic behavior. The rise time demonstrates the speed of response, while the overshoot highlights the
degree of initial overcompensation by the controller. The absence of significant undershoot and the
relatively quick settling time indicate that the system performs efficiently in reaching and maintaining the
desired state with minimal oscillations. Understanding these parameters is essential for evaluating the
control system's effectiveness in maintaining stability and meeting performance requirements.
(iii) Modified Block Diagram
Figure 4- Modified Block Diagram using Ramp.
Results:
Figure 5- Output response change with time.
(iv) Difference between the output vs. input trend obtained from the modified block diagram in b(iii) with
the output vs. input trend from the raw input and output data obtained from the Excel file.
Figure 6- Input and Output trend comparison.
The provided figures illustrate the comparison between the input trend and output trend obtained from
the modified Simulink block diagram and the raw data provided in the Excel file. Based on the analysis,
the following observations and conclusions are made:
Input Trend Comparison
The input trend from the Simulink model (blue dashed line) closely aligns with the raw input data from
the Excel file (red solid line). Both trends exhibit a stepwise behavior over time, where increases in input
flowrate are consistent in both datasets. This strong correlation indicates that the input signal in the
Simulink model has been effectively replicated from the real system data. Any minor discrepancies
between the two trends, though negligible, may arise due to differences in sampling frequency between
the Simulink simulation and the recorded Excel data or due to interpolation errors during data export or
plotting.
Output Trend Comparison
Similarly, the output trend from the Simulink model (blue dashed line) closely follows the pattern of the
raw output data from the Excel file (red solid line). Both datasets display stepwise increases in the output
temperature over time, indicating a consistent relationship between the input and output variables. This
alignment suggests that the transfer function used in the Simulink block diagram effectively captures the
dynamic behavior of the real system. Minor deviations, if present, are likely the result of simplifications
in the transfer function model or the presence of noise and disturbances in the actual recorded data that
were not accounted for in the Simulink simulation.
Input-Output Relationship
The input-output relationship is consistent across both datasets. Increases in input flowrate lead to
corresponding increases in output temperature, which stabilizes at each step. This indicates that the
Simulink model successfully replicates the system's dynamic response to changes in input conditions. The
close match in trends validates the accuracy of the transfer function and the block diagram in modeling
the actual system. Any minor differences can be attributed to approximations made during the system
identification process or measurement inaccuracies in the raw data.
Conclusion
Overall, the comparison between the input and output trends from the Simulink model and the Excel data
demonstrates a strong correlation, affirming the validity of the Simulink model as a representation of the
real system. The transfer function identified in part (a) effectively captures the system dynamics, as
evidenced by the close alignment of trends in both datasets. Minor discrepancies are negligible and are
likely due to simplifications in the modeling process or noise in the recorded data. This analysis confirms
the reliability of the Simulink model in predicting the system's behavior under varying input conditions.
Further refinement of the model, such as incorporating additional dynamics or noise components, could
be considered to enhance accuracy if necessary.
C. CONTROLLER DESIGN AND TUNING - AFIQAH AZZAHRA BINTI ZULKAFLY
(i) Block diagram of feedback control system.
Assuming the gain is 1 and the output response is of zeroth order for both the transmitter and the
control valve.
Figure 7- Block diagram of feedback control system.
Figure 8- Output response change with time using a feedback control system.
The results from the simulation, which show the Process Variable (PV) over time, provide some valuable
insights into how the system responds to a step change in the setpoint. By looking closely at the behavior
of the system, we can better understand why the results turned out the way they did.
At the start of the simulation, the Process Variable (PV) remains steady at about 0.011, which is expected
as the system is in a stable state before the setpoint is changed. This is typical behavior for control
systems, where the process variable stays constant until a new input is applied. Once the setpoint is
changed via the Setpoint (SP) block, the system tries to adjust the output to match the new target.
However, what stands out in the results is a sharp spike in the Process Variable (PV) at around 15,000
seconds. This sudden jump indicates a significant overshoot in the system’s response. This behavior can
be attributed to a few factors, especially the parameters set in the Transfer Function block, which
influences how the system responds to changes.
The transfer function used here, represented as 552 / s, models a first-order system with a proportional
gain of 552. The high gain is likely the reason for the system's strong reaction to the setpoint change.
When the gain is too high, the system becomes overly sensitive to even small differences between the
setpoint and the process variable. This leads to an exaggerated response, like the sharp spike we see in the
results. Ideally, the system would adjust more smoothly, with the process variable gradually approaching
the setpoint without sudden fluctuations.
Another reason for the overshoot is the lack of damping in the system. Damping is what helps smooth out
the response and prevent overshoot or oscillations. Without enough damping, the system doesn’t “settle
down” quickly enough, and we end up with spikes or oscillations after a change in input. The transfer
function used in the model does not include any damping, which contributes to the system's unstable
behavior.
There is also the possibility of numerical issues or simulation artifacts, which can occur if the solver
struggles to handle the system dynamics correctly, especially when high gain values are involved or if the
time steps aren’t small enough to capture the system’s behavior accurately.
To sum up, the sharp spike we see in the process variable is likely due to the high gain in the transfer
function, which makes the system too sensitive. The lack of damping makes it harder for the system to
stabilize smoothly, resulting in an overshoot. To improve the system's response, it would be helpful to
lower the gain and introduce damping into the transfer function. Fine-tuning these parameters will allow
the system to settle more gradually and avoid the sudden spikes, leading to a more stable and controlled
response.
(ii) Performance of the output response when the process input changes by a unit step by tuning the
controller parameters P, I and D.
Figure 9- Closed-Loop System Block Diagram for Process Control with Different PID Tunings.
Figure 10- Comparison of Output Response for Different PID Tuning Parameters in Closed-Loop System.
The graph illustrates the closed-loop response of the system to a unit step change in the process input
under three different PID parameter settings. By analyzing the rise time, overshoot, settling time, and
steady-state accuracy, the performance of each tuning mode is evaluated to identify the most optimal
configuration. This selection considers both the controller's cost and its suitability for chemical process
plant operations.
The first configuration, PID (P = 10, I = 1, D = 0), produces a slow response characterized by a long rise
time. Although there is no overshoot, the response is relatively sluggish, taking longer to reach the
steady-state value. This configuration minimizes computational cost due to its simplicity, as it lacks a
derivative term and has a low proportional gain. However, its slow response may not meet the
performance requirements of dynamic chemical processes where faster adjustments are critical.
In contrast, the second configuration, PID (P = 20, I = 2, D = 1), demonstrates a faster rise time with a
slight overshoot. The system stabilizes quickly, achieving steady-state conditions efficiently. The
inclusion of derivative action reduces oscillations and improves stability. While this tuning strikes a better
balance between speed and accuracy, its higher gain values and more complex controller design increase
computational cost, which may be a concern in large-scale operations.
The third configuration, PID (P = 15, I = 5, D = 0.5), achieves the best overall performance. It delivers a
fast rise time without significant overshoot and stabilizes quickly at the steady-state value. The higher
integral term effectively eliminates steady-state error, while the moderate derivative term ensures a
smooth response by reducing oscillations. This configuration provides a good trade-off between
performance and cost, making it both efficient and cost-effective for implementation.
For applications in a chemical process plant, the third configuration is the most suitable. Chemical
processes often require controllers that can respond quickly and accurately to maintain critical variables,
such as temperature, pressure, or flow rate, within narrow operational limits. Overshoot or oscillations can
pose safety risks, affect product quality, and lead to equipment damage. Therefore, a configuration that
balances speed, stability, and computational efficiency is essential.
In conclusion, the third tuning mode, PID (P = 15, I = 5, D = 0.5), is the most optimal for the given
system. It combines fast and stable performance with moderate computational cost, aligning well with the
requirements of a chemical process plant. By ensuring responsive, stable, and cost-efficient operation, this
tuning mode provides an effective solution for maintaining control in dynamic industrial environments.
REFERENCES
[1] Introduction to system Identification. (n.d.). FIRST Robotics Competition Documentation.
[Link]
APPENDICES
1st order system:
2nd order system:
Highest order system:
