An Empirical Approach in Data-driven PID Controller Tuning of Temperature Control Loop in Distributed Control System (DCS) of a Petroleum Refinery Industry

[1]¿p#1 This study presents a systematic methodology for the preliminary tuning of a proportional–integral–derivative (PID) controller implemented within a distributed control system (DCS) for regulating the bottom temperature of a stabilizer column in a catalytic reforming unit of a petroleum refinery industry controlled by TIC-3051-1. To enable stable automatic operation, real-time historical process plant data comprising approximately 300 records of process and manipulated variables were extracted for 17 minutes from the DeltaV DCS panel. The transfer function tf2 of the process was identified through MATLAB system identification techniques. The developed model was subsequently integrated into a Simulink-based control framework. Automatic PID tuning was then performed to obtain appropriate proportional (Kp), integral (Ki), and derivative (Kd) gains, which were estimated as 0.611, 0.022, and 4.16, respectively, and further, these were converted into DCS-compatible controller parameters as Kc = 0.63, Ti = 26 seconds, and Td = 6.4 seconds. 14 hours of operational trends demonstrated that the tuned controller maintained the set-point temperature of 238.4℃ within an acceptable range over extended operation. The results confirm that data-driven preliminary tuning using MATLAB–Simulink provides an effective and practical approach for improving temperature control performance in refinery DCS applications. Md Ryshur Rahman Turin, Nasim Ahmed Saeed, and Md. Adnan Faisal Siddique 1 Institute of Energy Technology, Chittagong University of Engineering and Technology, Chattogram-4349, Bangladesh 2 Department of Chemical Engineering and Polymer Science, Shahjalal University of Science and Technology, Sylhet-3114, Bangladesh * Corresponding Author: [email protected]

This study presents a systematic methodology for the preliminary tuning of a proportional–integral–derivative (PID) controller implemented within a distributed control system (DCS) for regulating the bottom temperature of a stabilizer column in a catalytic reforming unit of a petroleum refinery industry controlled by TIC-3051-1. To enable stable automatic operation, real-time historical process plant data comprising approximately 300 records of process and manipulated variables were extracted for 17 minutes from the DeltaV DCS panel. The transfer function tf2 of the process was identified through MATLAB system identification techniques. The developed model was subsequently integrated into a Simulink-based control framework. Automatic PID tuning was then performed to obtain appropriate proportional (Kp), integral (Ki), and derivative (Kd) gains, which were estimated as 0.611, 0.022, and 4.16, respectively, and further, these were converted into DCS-compatible controller parameters as Kc = 0.63, Ti = 26 seconds, and Td = 6.4 seconds. 14 hours of operational trends demonstrated that the tuned controller maintained the set-point temperature of 238.4℃ within an acceptable range over extended operation. The results confirm that data-driven preliminary tuning using MATLAB–Simulink provides an effective and practical approach for improving temperature control performance in refinery DCS applications. Keywords: PI Control, Temperature Control, Tuning, Distributed Control, Controllers

Temperature regulation within petroleum refinery units constitutes one of the most critical operational requirements, as deviations from prescribed thermal conditions directly affect product quality, separation efficiency, energy consumption, and overall plant safety. In modern refineries, these regulatory tasks are predominantly executed through Distributed Control Systems (DCS) employing Proportional–Integral–Derivative (PID) controllers because of their structural simplicity, ease of implementation, and proven robustness under industrial disturbances [1]. Despite their widespread adoption, conventional PID loops often suffer from suboptimal tuning, leading to oscillatory responses, excessive settling times, and energy inefficiencies, particularly in highly nonlinear and integrating thermal processes typical of furnaces, heat exchangers, and distillation columns. Recent studies have demonstrated that data-driven controller design can achieve comparable or superior performance relative to classical approaches while significantly reducing modeling complexity [2]. Within the context of petroleum refining, several investigations have emphasized the necessity of systematic performance monitoring and diagnosis of control loops to mitigate oscillations and inefficiencies [3]. Applications of advanced PID and nonlinear control schemes to refinery units such as fluid catalytic cracking and heating furnaces have demonstrated notable improvements in temperature stability and fuel efficiency [4]. Self-tuning and adaptive controllers have also been successfully implemented to compensate for process variability, thereby enhancing operational reliability [5]. Consequently, model-based and Internal Model Control (IMC)–based tuning frameworks have been introduced to enhance stability and robustness in process industries, especially in atmospheric and vacuum distillation systems [6]. Nevertheless, the requirement for accurate plant models often limits their practicality in large-scale industrial environments where process dynamics are uncertain or time-varying. Database-driven updating strategies, adaptive retuning algorithms, and closed-loop identification techniques have further enabled continuous improvement of PID parameters directly from plant measurements [7]. These strategies are particularly advantageous for refinery processes characterized by nonlinearities, disturbances, and frequent throughput variations. Furthermore, optimization-based and evolutionary tuning methods have been explored to automatically determine near-optimal parameters for complex thermal systems [8]. More recently, intelligent and machine-learning-assisted PID frameworks, including physics-informed and neural network–based approaches, have provided promising avenues for adaptive and predictive control of industrial processes [9]. Although these advanced techniques offer improved adaptability, their computational and implementation complexities often hinder immediate deployment within existing DCS infrastructures, where simplicity, transparency, and operator interpretability remain essential. Historically, heuristic and empirical tuning strategies such as the Ziegler–Nichols’s method have been extensively utilized for preliminary controller parameterization [10]; however, these approaches frequently fail to ensure satisfactory performance under varying refinery operating conditions. The optimization of the fractionation unit constitutes a considerable engineering challenge, owing to the inherent complexity of diverse refinery configurations—including intricate tray geometries and internal arrangements—as well as persistent operational constraints such as flooding, weeping, and associated hydraulic instabilities [11]. In this regard, an empirical yet data-driven preliminary tuning methodology that integrates real plant measurements with classical PID structure offers a practical compromise between simplicity and performance. Such an approach retains the familiarity of conventional PID controllers while leveraging historical process data to systematically refine control parameters, thereby ensuring direct applicability within refinery DCS platforms. Therefore, this study proposes an empirical data-driven framework for preliminary PID tuning of a temperature control loop in the petroleum refinery industry, aiming to bridge the gap between theoretical controller design and practical industrial implementation while improving stability, responsiveness, and energy efficiency. 2. Materials and Methods PID controller parameters were optimized within the simulation procedure to determine the proportional gain, integral reset time, and derivative rate. Both classical empirical correlations and iterative performance-based tuning were applied, using criteria such as settling time, overshoot, and integral error metrics to achieve a balanced trade-off between responsiveness and robustness. Such structured tuning methodologies are particularly effective for integrating and delay-dominant refinery processes [1]. The proposed empirical, data-driven PID tuning framework was initiated through systematic extraction of operational measurements from the Distributed Control System (DCS) record sheets of an operating petroleum refinery. Time-series records of heater bottom temperature and corresponding control valve opening were exported in spreadsheet format over representative steady-state and transient regimes to ensure adequate excitation of process dynamics. The use of process plant data provides a realistic representation of nonlinearities, disturbances, and actuator limitations inherent in petroleum refinery operations, thereby offering a reliable basis for controller performance assessment and retuning under true industrial conditions [3]. The acquired datasets were subsequently imported into the MATLAB computational environment for signal conditioning and numerical analysis. Pre-processing steps, including noise attenuation, outlier rejection, time alignment, detrending, and normalization, were conducted to improve data fidelity and statistical consistency prior to modeling. MATLAB-based processing facilitates efficient manipulation of large industrial datasets and supports advanced identification and estimation algorithms suitable for data-driven control development without requiring explicit first-principles models [12]. A dynamic model of the temperature control loop was then derived via transfer function estimation using MATLAB’s System Identification Toolbox. Treating valve position as the manipulated input and temperature as the controlled output, the process behavior was approximated using a low-order structure, typically represented by a first- or second-order plus dead-time (FOPDT/SOPDT) formulation. Such parametric models capture the dominant gain, time constant, and delay characteristics governing thermal process dynamics and provide an adequate mathematical abstraction for subsequent controller synthesis [13]. Based on the identified model, a closed-loop simulation architecture was constructed within the Simulink platform to replicate the dynamic interaction among the process, measurement elements, and control actions. The block diagram enabled virtual prototyping of the refinery temperature loop, allowing safe evaluation of multiple control configurations and transient scenarios without disturbing plant operation. This model-based simulation approach is widely adopted for pre-implementation validation of industrial control strategies [14]. The estimated numerator and denominator coefficients obtained from the identification stage were directly embedded into the transfer function block to parameterize the virtual plant. This step ensured consistency between empirical process behavior and simulated dynamics, thereby enhancing the predictive accuracy of the model and enabling reliable controller performance testing under realistic conditions [6]. Subsequently, after achieving satisfactory closed-loop performance in simulation, the optimized parameters were implemented directly in the DCS system of the Catalytic Reforming Unit. The calculated gain, reset, and rate values were configured in the PID controller to replicate the validated simulation response, thereby minimizing commissioning effort and enabling rapid deployment of improved loop settings in the operating unit [7]. Finally, the effectiveness of the proposed strategy was validated through on-site monitoring of temperature trajectories and valve opening profiles during normal operation. Comparative analysis against historical baseline data confirmed reductions in oscillation amplitude, improved set-point tracking, and enhanced process stability. This real-time verification demonstrates that the data-driven simulation-assisted tuning framework provides a practical and transferable methodology for refinery temperature control loop [15]. 2.1 Schematic Overview of PID Controller In order to run a PID in auto mode, preliminary tuning needs to be done. In this particular case, a temperature-indicating controller (TIC) has been tuned using MATLAB and Simulink. The controller is used to control the bottom temperature of a stabilizer column of the catalytic reforming unit of a Natural Gas Condensate (NGC) refinery. It is done by controlling the reboiling outlet temperature of a fired heater. In the controller the manipulated value (MV) is the valve opening for the fuel supply to the fired heater, and the process value (PV) is the heater outlet temperature. Fig. 1 A typical overview of PID Controller TIC-3051-1 in Catalytic Reforming Unit Figure 1 shows a schematic representation of the temperature-indicating controller (TIC-3051-1) within the catalytic reforming unit illustrating the closed-loop configuration between the fired heater outlet temperature and the fuel control valve, thereby establishing the manipulated–controlled variable relationship that forms the basis of subsequent PID tuning and automation. TIC-3051-1, the controller that is used to control the reboiling temperature of the fired heater H-304. Running it manually and controlling the fuel valve opening has many drawbacks since the temperature is manipulated by many factors or disturbances, such as fuel quality, fuel pressure after the regulator, column liquid level, column pressure, air flow of the blower, etc. So, running it automatically would be much more convenient. But the controller is not tuned yet; it’ll not be able to properly control the temperature. In order to avoid overshoot and turbulent response, it needs to be tuned on a preliminary basis, which can be done using MATLAB and Simulink. 2.2 Extracting the data Establishing the transfer function of the process is the first step of tuning. The easiest and most convenient way is to extract the real-time data of PV and MV and estimate the transfer function using MATLAB. The DCS software (Delta V) provided by Emerson has a built-in add-in option with Microsoft Excel, which can be used to collect the historical data from trends of different transmitters and controllers. In this case 300 data of the temperature and fuel valve opening have been taken in the span of around 17 minutes. As the data sets are interpolated in nature, the missing data points of the trend line are compensated using interpolation. Fig. 2 Temperature profile and valve opening data taken from DCS Software of Emerson Figure 2 depicts historical operational data acquired from the Emerson DeltaV distributed control system showing synchronized temperature and valve-opening trends over 17 minutes, which serve as real-plant input–output signals for empirical system identification and controller design. Fig. 3 Inclusion of temperature profile and valve opening in MATLAB Figure 3 highlights imported process variables within the MATLAB workspace, demonstrating the integration of manipulated variables as mv = data.mv and process variables as pv = data.pv into a computational system for preprocessing, visualization, and modeling. 2.3 Estimating the transfer function A simple code can import the Excel data to MATLAB. The data of PV and MV are then used to estimate the transfer function using the System Identification Chapter of MATLAB. Figure 3 highlights imported process variables within the MATLAB workspace, demonstrating the integration of measured temperature and actuator position data into a computational environment for preprocessing, visualization, and dynamic modeling. Fig. 4 Transfer Function was generated in MATLAB utilizing the input data Transfer function estimation results are obtained as tf1 and tf2 using the MATLAB System Identification Toolbox, where the refinery heater dynamics are approximated by a low-order model capturing dominant gain, time constant, and oscillatory characteristics, as shown in figure 4. Taking poles=2 and zeros=0, as the data shows an oscillatory response, the transfer function tf2 is exported to the MATLAB workspace. 2.4 Setting up Simulink blocks A step response is fed to the Simulink loop, and a delay of 5 seconds is used for overall system delay. Fig. 5 Simulation blocks were set by Simulink The Simulink-based closed-loop structure in figure 5, depicting the virtual emulation of plant dynamics, control action, and feedback pathways, provides the numerator and denominator of transfer function tf2 prior to field implementation. Fig. 6 The numerator and denominator values are inserted in the transfer function block Figure 6 represents parameterization of the transfer function block through insertion of the identified numerator and denominator coefficients to ensure consistency between empirical plant behavior and the simulated process response. The response of the control loop can be observed by running the program. 3. Results and Discussion The effectiveness of the proposed data-driven PID tuning approach was assessed through both simulation and plant-level validation. Time-domain amplitude–time responses were first analyzed to evaluate transient performance, including rise time, overshoot, and settling behavior of the tuned loop. The optimized proportional, integral, and derivative gains were subsequently obtained from the PID block parameters and converted into industrial control constants \((K_{c},\ T_{i},\ T_{d})\)for direct implementation within the Distributed Control System (DCS). Fig. 7 Reference tracking of tuned versus block response 3.1 Tuning the PID Controller The automatic tuning option in the PID block gives an option to tune the PID according to the desired response. Figure 7 illustrates a comparative reference-tracking response demonstrating improved transient characteristics of the tuned controller relative to the untuned configuration, highlighting reductions in overshoot and settling time within the simulation framework. Updating the block gives the values of proportional, integral, and derivative gains (Kp, Ki, Kd). Fig. 8 The obtained values of proportional, integral, and derivative gains (Kp, Ki, Kd) Figure 8 shows extracted proportional, integral, and derivative gain values (Kp, Ki, Kd) computed from the automatic tuning algorithm, representing the optimized control parameters required for stable refinery temperature regulation. The proportional gain (\(K_{c}\)), integral reset time (\(T_{i}\)), and derivative rate time (\(T_{d}\)) collectively regulate the responsiveness, steady-state accuracy, and damping characteristics of the refinery temperature loop, where\(K_{c}\)determines control aggressiveness, integral action eliminates offset, and derivative action enhances stability by anticipating process trends. Converting the Simulink parameters to DCS parameters by using • Controller gain = \(K_{c}\) • Proportional gain = \(K_{p}\) • Integral gain (reset action) = \(K_{i}\) • Derivative gain (rate action) = \(K_{d}\) In most DCS systems, controller gain and proportional gain are the same parameters. Thus, Kc = Kp Additionally, in DCS systems, reset time \(T_{i}\ \)is often used, and thus the relationship between \(T_{i}\ \)and \(K_{i}\) is \begin{equation} T_{i}=\ \frac{K_{p}}{K_{i}}\nonumber \\ \end{equation} Also, in DCS systems, derivative time \(T_{d}\) is mostly used, and thus the relationship between \(T_{d}\) and \(K_{d}\ \)is, \begin{equation} T_{d}=\ \frac{K_{d}}{K_{p}}\nonumber \\ \end{equation} Calculating the DCS parameters, we get Kc = Kp = 0.611 For estimating reset time, we get\(T_{i}=\ \frac{K_{p}}{K_{i}}=\ \frac{0.611}{0.022}=27.8\ sec\) For determining rate time, we get\(T_{d}=\ \frac{K_{d}}{K_{p}}=\ \frac{4.16}{0.611}=6.8\ sec\) The most suitable PID parameters were systematically identified using an iterative optimization approach to obtain improved stability, reduced overshoot, and faster settling characteristics: \begin{equation} K_{c}=0.63\nonumber \\ \end{equation}\begin{equation} T_{i}=26\ sec\nonumber \\ \end{equation}\begin{equation} T_{d}=6.4\ sec\nonumber \\ \end{equation} The iteratively optimized PID parameters were subsequently implemented within the Distributed Control System (DCS) to validate their performance under real plant operating conditions. Fig. 9 Tuning proportional, integral, and derivative gains (Kp, Ki, Kd) in DCS System Implementation interface of the tuned PID parameters within the DCS platform illustrating the conversion of simulation-derived gains into DCS system gain, reset, and rate settings for direct deployment in plant operation as shown in figure 9. 3.2 The trend of temperature profile and valve opening after tuning Running the controller in auto mode during the operation gives the result as illustrated in figure 10. It represents an extended-duration operational temperature profile over a 14-hour period under automatic control mode, demonstrating sustained set-point tracking and reduced oscillatory behavior despite external disturbances. Fig. 10 The temperature profile in between the 14-hour time schedule Here, the set point is 238.4°C; the trend of temperature in the span of 14 hours shows a moderately consistent result. Even if the temperature stays in the desired range of value, because of the changing fuel pressure after the regulator, the valve opening of the fuel may rise or fall significantly. The ever-changing pressure of fuel after the regulator doesn’t let temperature have a stable value. Although the oscillatory response can be minimized if the after-regulator pressure is constantly monitored and controlled manually if necessary. Fig. 11 The trend of valve opening and temperature variation after tuning Figure 11 highlights the post-tuning trend comparison of valve opening and temperature responses for about 105 minutes duration, showing consistency between both the profiles after tuning, thereby validating the practical effectiveness of the data-driven PID controller tuning in real refinery conditions. 4. Conclusion The present investigation successfully demonstrates that preliminary PID tuning based on real plant data and simulation-based analysis can significantly enhance temperature control performance in refinery DCS applications. By extracting the preceding 300 real-time data and estimating an accurate dynamic transfer function as tf2, the inherent oscillatory behavior of the fired heater temperature control loop was effectively captured and addressed. The implementation of MATLAB for generating tf2 and Simulink for yielding simulation blocks enabled the determination of suitable PID parameters, Kc, Ti, and Td, which, when deployed in the DCS, resulted in stable and consistent temperature regulation around the desired set point of approximately 238.4℃ over 14 hours duration. Although variations in downstream fuel pressure continued to influence valve positioning, the tuned controller mitigated excessive oscillations and prevented unstable responses. Overall, the study confirms that preliminary PID tuning using simulation-based tools offers a reliable, low-risk strategy for transitioning critical temperature control loops from manual to automatic operation, thereby improving operational stability, safety, and process efficiency. Declarations • Ethics statement and consent to participate: Not applicable. This study did not involve any human or animal participants. • Consent for Publication: Not applicable. • Availability of Data and Material: All relevant data and materials are provided in the main manuscript. • Conflict of interest disclosures: The authors have no relevant financial or non-financial interests to disclose. • Funding: The authors declare that no funds, grants, or other support were received during the preparation of this manuscript. • Authors’ Contributions: All authors contributed to the study conception and manuscript preparation. Md Ryshur Rahman Turin conceived the study, carried out the experimental work, performed data analysis, and drafted the manuscript. Nasim Ahmed Saeed assisted with DCS operation, MATLAB, and Simulink. Md. Adnan Faisal Siddique supervised the research and critically reviewed the manuscript. All authors read and approved the final manuscript. • Acknowledgment: We are deeply grateful to the esteemed authorities of Partex Petro Limited, Chattogram, Bangladesh, for their continued encouragement and institutional support.

1. [1] Åström, K. J., & Hägglund, T. “ PID Controllers: Theory, Design, and Tuning .” ISA—The Instrumentation, Systems, and Automation Society, 1995. pp. 343. 2. [2] C.J. Munaro, M.R. Pimentel et al., “Data-Driven Performance Monitoring and Retuning using PID Controllers,” Computers and Chemical Engineering, October 2023, Volume 178, Issue 108360. https://doi.org/10.1016/j.compchemeng.2023.108360. 3. [3] Thornhill, N. F.; Oettinger, M.; Fedenczuk, P. “Performance Assessment and Diagnosis of Refinery Control Loops.” AIChE Symposium Series No 320, 1998, Volume 94, pp. 373-379 . 4. [4] Araromi, Dauda Olurotimi. “Feedback Control of Fluid Catalytic Cracking Unit Using Proportional Integral (PI) Controller.” International Journal of Engineering Research and Technology, 2015. 5. [5] S. Kotob et al., “Analysis and applications of self-tuning controls in a refining process: Case study.” Automatica, 1994, Volume 30, Issue 11, pp. 1663-1675. https://doi.org/10.1016/0005-1098(94)90070-1. 6. [6] Skogestad, S. “Simple Analytic Rules for Model Reduction and PID Tuning.” Journal of Process Control 2003, Volume 13, Issue 4, pp. 291–309. https://doi.org/10.1016/S0959-1524(02)00062-8. 7. [7] Wakitani, S.; Yamamoto, T.; Gopaluni, B. “Design and Application of a Database-Driven PID Controller with Data-Driven Updating Algorithm.” Industrial and Engineering Chemistry Research, 2019, Volume 58, Issue 26, pp. 11412–11423. https://doi.org/10.1021/acs.iecr.9b00704. 8. [8] Liu, N., Chai, T., Zhang, Y., et al. “Data-driven optimal tuning of PID controller parameters.” Science China Information Sciences, 2025, Volume 68 (172201). https://doi.org/10.1007/s11432-024-4292-y. 9. [9] Kos, Tomaž, Mikuláš Huba, and Damir Vrančić. ”Parametric and Nonparametric PI Controller Tuning Method for Integrating Processes Based on Magnitude Optimum.” Applied Sciences, 2020, 10, no. 4: 1443. https://doi.org/10.3390/app10041443. 10. [10] Ziegler, J.G. and Nichols, N.B., “Optimum Settings for Automatic Controllers.” Transactions of the ASME, 1942, 64, pp. 759-768. 11. [11] Md. Ryshur Rahman Turin, Md. Moniruzzaman, Anisul Islam Suva et al. “A Feasibility Study of Condensate Fractionation Unit (CFU) Process Using Aspen HYSYS Simulation Software in the Petroleum Refinery Industry of Bangladesh.” 2025 (Version 1), Research Square. https://doi.org/10.21203/rs.3.rs-6495903/v1. 12. [12] Bazanella, A. S.; Campestrini, L.; Eckhard, D., “ Data-Driven Controller Design: The H₂ Approach .” Springer, 2012. 13. [13] Ljung, L., “System identification: theory for the user.” 2nd edition; Prentice Hall, 1999. 14. [14] Model-Based Design with Simulink: Dynamic system simulation for model-based design, MathWorks Documentation, 2023. 15. [15] Biao Huang, Sirish L. Shah, “Performance Assessment of Control Loops: Theory and Applications.” 1 st edition, Springer, 1999. Information & Authors Information Version history Copyright This work is licensed under a Non Exclusive No Reuse License.

Authors Metrics & Citations Metrics Article Usage 176views 81downloads Citations Download citation Md Ryshur Rahman Turin, Nasim Ahmed Saeed, Md. Adnan Siddique. An Empirical Approach in Data-driven PID Controller Tuning of Temperature Control Loop in Distributed Control System (DCS) of a Petroleum Refinery Industry. Authorea. 23 March 2026. DOI: https://doi.org/10.22541/au.177425134.40079380/v1 DOI: https://doi.org/10.22541/au.177425134.40079380/v1 If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download. For more information or tips please see 'Downloading to a citation manager' in the Help menu.