Abstract
This paper discusses the plant-wide control system of a biohydrogen plant from biomass. The plant’s control structure is designed by assessing the use of model predictive control (MPC) and proportional-integral (PI) controllers for each controlled variable. Then, the control structure design is tested by set point and disturbance change tests, and its performance is evaluated by integral square error (ISE). This plant is simulated in two different capacities to understand the effect of change in plant capacity on the plant’s controllability. Results show that the plant with capacity A has controllability of +25% and −5%, while the plant with capacity B has controllability of +100% and −5% for the dry biomass molar flow rate change test. Both plants have the same controllability of +100% and −100% for the dry biomass temperature test. Both plants result in a dry biomass conversion rate of 16.52%, where the amount of H2 produced is 382 kg/h and 2,291 kg/h, respectively.
1. Introduction
Hydrogen is predicted to be the future of clean energy, and it is crucial to produce hydrogen cleanly and sustainably. Hydrogen is very beneficial for human life. In organic chemistry, it is used for hydrogenation reactions such as the Rosenmund chemical reaction [1] and the Friedel–Crafts acylation chemical reaction [2]. Hydrogen is also used in the fertilizer, food, cooking oil, and fuel industries [3]. Its use in fuel cells is also very prominent [4].
There are various ways to produce hydrogen, but there is an increasing interest in utilizing biomass due to its resource diversity, availability, and sustainability [5, 6]. Three routes of producing biohydrogen from biomass are thermochemical, biological, and electrochemical processes [7]. Three types of thermochemical processes are gasification, pyrolysis, and aqueous phase reforming (APR) [5, 7]. Five types of biological processes are direct photolysis, indirect photolysis, biological water gas shift (BWGS) reaction, photofermentation, and dark fermentation [7, 8]. Meanwhile, two types of electrochemical processes are proton exchange membrane electrolysis cell (PEMEC) and microbial electrolysis cell (MEC) [7]. The factors considered in selecting the technology are but are not limited to H2 yield, technology readiness level (TRL), and economic feasibility [9]. In this work, the gasification process is chosen due to its high energy recovery and efficiency [6, 10], highest H2 yield [5, 10], best technological maturity [7], and it is the most economically feasible process to implement on an industrial scale [5, 8].
Despite the feasibility of the gasification process to produce biohydrogen, more research needs to be done before implementing it on an industrial scale. One of the urgent issues to be addressed is the scalability and controllability of the biohydrogen production process [10]. Yuan et al. [11] defined controllability as how easy it is to control a plant within an acceptable operating range despite any disturbance in the system. Controllability is a crucial issue to consider when scaling up a pilot scale plant into an industrial scale plant to ensure a smooth production process and stability and reliability of the H2 supply. Plant controllability analysis is conducted using integral square error (ISE) to evaluate the controllers’ performance in controlling the plant, i.e., handling set point and disturbance changes in the system. Currently, no research is conducted on analyzing the effect of change in plant capacity on plant controllability. Therefore, this research will evaluate the controllers’ performance in two different plant capacities: capacity A with 2,312 kg/h of dry biomass and capacity B with 13,870 kg/h of dry biomass. The results will be analyzed to see whether the change in plant capacity will affect plant controllability and to what extent the difference in the plant controllability between the plant with capacities A and B.
Before assessing the plant’s controllability, the plant-wide control system must be designed first. Previous works have applied process control to the biohydrogen plant from biomass, where each unit operation is controlled and optimized separately. Salma [12] used proportional-integral (PI) controllers for the gasifier, char combustor, and flue gas cooler. Wahid and Nararya [13] improved the control performance by reidentifying the system models and applying model predictive control (MPC) for the gasifier, char combustor, and flue gas cooler. Wahid and Iqbal [14] used PI controllers for the compressor and steam methane reformer (SMR). Wahid and Taqwallah [15] improved the control performance by reidentifying the system models and applying MPC for the compressor and SMR, and Adjisetya and Wahid [16] used multivariable model predictive control (MMPC) for the compressor and steam methane reformer (SMR) control. This research will continue developing the simulation by combining all unit operations into one simulation and assigning each controlled variable with either an MPC or PI controller. Then, the system models are identified, and the controllers are tuned to achieve optimum control performance. Afsi et al. [17] have also researched the use of PI and MPC in bioprocesses, namely, polylactic acid (PLA), as well as using dynamic optimization, and then chose which one is better. In contrast to this research, the research did not choose which was the best of them but chose PI and MPC based on the accuracy of using both in controlling certain variables.
Based on the explanation above, the discussion in this paper will be arranged as follows. Section 2 discusses the steady-state simulation of the plant, consisting of the process description and process flowsheet in UniSim Design R491. Section 3 discusses the dynamic simulation of the plant, consisting of the plant’s control structure design, system identification, controller tuning, controller testing, performance, and plant controllability. Section 4 discusses the conclusion and future directions of this research.
2. Materials and Methods
2.1. Steady State Simulation
Figure 1 shows the steady-state simulation of the biohydrogen plant from biomass. The plant design is based on the work by Budianta et al. [9] with modifications. The feed to the biohydrogen plant is assumed to be dry biomass containing 92.77% carbon and 7.23% water (i.e., the biomass has been pretreated, so there is no pretreatment equipment in this simulation). Dry biomass enters directly into the gasifier together with steam, where it is converted into gases such as CO, CO2, H2O, H2, and CH4. The amount of steam used is based on the steam to biomass ratio of 0.75 [18]. The gasifier is simulated as a conversion reactor in UniSim, with reactions occurring shown in equations (1)–(4) [19]. The outlet streams of the gasifier are divided into two; the “gas” stream flows to the SMR, while the “char” stream flows to the char combustor.
The “char” stream flows to the char combustor, where it is reacted with air to produce heat for the gasification process. The char combustor is simulated as a conversion reactor in UniSim, with reactions occurring shown in equations (5) and (6). The outlet streams of the char combustor are divided into two; the “flue gas” stream flows to the flue gas cooler before being discarded, while the “ash” stream is discarded.
The “gas” stream flows to the SMR, where it is reacted with steam to convert CH4 into CO and more H2. The SMR is simulated as a plug flow reactor in UniSim, with the reaction occurring shown in equation (7). The outlet stream, namely, “gas out SMR,” flows to the cooler to lower the temperature before entering the water gas shift (WGS) reactor together with more steam to convert CO into CO2 and more H2. The WGS reactor is simulated as a conversion reactor in UniSim with reactions occurring shown in equation (8). The outlet stream of the WGS reactor is the final product, namely, “H2 gas,” which contains 61.95% H2, 30.96% CO2, and traces amount of CO and H2O.
This plant is simulated in two capacities, capacity A with 2,312 kg/h of dry biomass and capacity B with 13,870 kg/h of dry biomass, which aims to mimic the pilot scale and industrial scale plant. The amount of H2 produced in capacity A is 382 kg/h, while in capacity B is 2,291 kg/h. Both plants with capacities A and B result in a dry biomass conversion rate of 16.52%. This capacity is calculated based on Indonesia’s estimated hydrogen imports in 2040, which amount to 908.5 kg/h. Therefore, capacity A accounts for 42% of the total hydrogen imports, while capacity B is planned to substitute all hydrogen imports and the remainder will be for export.
2.2. Dynamic Simulation
2.2.1. Control Structure Design
Figure 2 shows the dynamic simulation of the biohydrogen plant from biomass with controllers. There are five pairs of controlled and manipulated variables, as shown in Table 1. Temperature is the control focus of this plant because the temperature in each unit operation needs to be controlled to achieve the maximum reaction conversion, which correlates to the final amount of H2 produced.
In this simulation, we consider using both MPC and PI controllers, and therefore, an assessment is conducted to determine the controller used for each controlled variable. The two factors considered in this assessment are (1) the controller’s performance in controlling the system and (2) the urgency to apply a more complex controller to achieve the control objective. According to Kano and Ogawa [20], MPC outperforms the PI controller in almost every controller type (i.e., temperature, flow, pressure, and concentration) except for the level controller. However, not all unit operations directly affect the final amount of H2 produced, e.g., char combustor and flue gas cooler, and therefore, a PI controller is sufficient for these unit operations. Table 2 concludes the chosen controller type for each controlled variable.
2.2.2. System Identification and Controller Tuning
All systems in this plant are assumed to follow a first-order linear plus dead time (FOPDT) model, not a nonlinear model. This is intended to simplify the optimization problem. This low-level model captures the important dynamics of the system while reducing computational complexity. The mathematical formulation of predictive control is essentially based on linear models [21]. Linear models simplify optimization problems and enable the use of efficient numerical optimization techniques, such as quadratic programming, to solve optimization problems at any control interval. System identification is conducted to obtain the FOPDT models, as shown in Tables 3 and 4.
In MPC, the prediction (P) and control (M) horizons and sampling time (T) in MPC determine how far in the future the system is predicted and optimized. Longer horizons generally produce more accurate predictions but also increase computational complexity. This tuning is based on specific applications and system dynamics to find a balance between accuracy and computational efficiency. To tune the MPC parameters, the Shridhar and Cooper method [22] was first used and then continued with fine tuning to enhance the robustness of the MPC controller so that it can handle mismatches between the model and reality. Meanwhile, for the PI controller, the Ziegler–Nichols method [23] is used. This method is known for producing settings that tend to minimize oscillations and stabilize the system. The method aims to strike a balance between responsiveness and stability, making it a good starting point for many control systems. Although this method was basically developed to tune a PI controller to achieve the desired balance between responsiveness and stability when there is a setpoint change and does not have specific guidelines for tuning a PI controller to handle disturbances, however, integral action in the calculated PI parameters can help reduce errors in a steady state caused by disturbances over time. The integral term allows the controller to continuously adjust the control output to eliminate long-term offsets.
The tuning parameters used for both MPC and PI controllers are shown in Tables 3 and 4, respectively. Both plants with capacities A and B have the same controller type and tuning parameters.
3. Results and Discussion
Controller testing is conducted to evaluate the controllers’ performance in handling set point and disturbance changes. In the set point change test, the initial set point of each controlled variable is increased with a determined increment, as shown in Table 5. The test’s objective is to see whether the controller can adjust the manipulated variable to the new set point, therefore minimizing the difference between the new value of the set point and the controlled variable.
In the disturbance change test, the value of the chosen disturbance is changed from its initial value with a determined percentage. The two disturbances in this test are dry biomass molar flow rate and dry biomass temperature. The percentage change of the dry biomass molar flow rate ranges from +5% to +100% and −5% to −10%. Meanwhile, the percentage change of the dry biomass temperature ranges from +50% to +100% and −50% to −100%. The test’s objective is to see whether the controller can adjust the manipulated variable so that the controlled variable can return to its initial value, therefore minimizing the difference between the unchanged value of the set point and the controlled variable.
ISE is used to evaluate the controllers’ performance quantitatively. A low ISE value is desired, meaning that the error between the values of the set point and the controlled variable is minimum, and the controller can adjust quickly to the changes. The percentage determines the plant’s controllability, meaning to what extent the controllers can handle disturbance changes in the plant.
Table 5 shows the ISE values for the set point change test in both plants with capacities A and B. Figure 3 shows the graph results of the set point change test in both plants with capacities A and B. Overall, the controllers in both plants perform well. However, the ISE values of the plant with capacity B are larger than those of capacity A. This result means that a larger plant capacity is more challenging to control with the current controllers’ tuning parameters.
(a)
(b)
(c)
(d)
(e)
Tables 6 and 7 show the ISE values for the dry biomass molar flow rate change test in plants with capacities A and B, respectively. The controllers in the plant with capacity A can handle disturbance only up to +25% and −5% from its initial value. Meanwhile, the controllers in the plant with capacity B can handle disturbance up to +100% and −5% from its initial value. Although the controllers can handle disturbance in the plant with capacity B, the ISE values are larger than those of capacity A. This is due to the ISE equation, where the difference between the set point and the controlled variable is squared. With the same disturbance percentage, the disturbance value in the plant with capacity B is larger than that of capacity A. A larger disturbance means that the error is larger, and the error is also squared, therefore increasing the ISE value. Based on these results, increasing factory capacity must be accompanied by reidentifying the system because different capacities will affect the behavior of the processes within it. Next, the MPC and PI controller parameters must be readjusted so that they match the changed process behavior.
Tables 8 and 9 show the ISE values for the dry biomass temperature change test in plants with capacities A and B. Overall, the controllers in both plants perform well, where they can handle disturbance up to +100% and −100% from their initial value. The ISE values in both plants are also relatively smaller than those of the dry biomass molar flow rate change test, showing that the dry biomass temperature does not affect the system as much as the dry biomass molar flow rate.
4. Conclusion
This work proposed a plant-wide control structure design of a biohydrogen plant from biomass using a combination of MPC and PI controllers. MPC is used in the gasifier, SMR, and cooler before the WGS reactor, while the PI controller is used in the char combustor and flue gas cooler. Results show that with the same controller type and tuning parameters for both plants with capacities A and B, the controllers performed relatively well in handling set point and disturbance changes. In terms of the disturbance, the dry biomass molar flow rate affects the plant’s controllability more significantly compared to the dry biomass temperature. However, with higher ISE values of the plant with capacity B, it is advisable to reidentify the system and retune the controllers if the plant capacity changes to optimize their performance. Further work can be done to improve steady-state simulations by adding more detailed unit operations and improving controller performance by applying multivariable model predictive control (MMPC) to account for interactions between variables in the plant or using a hybrid approach that combines the strengths of MPC and PI controllers as has been researched by the author himself [24, 25].
Data Availability
The data used to support the findings of this study are available from the authors on the following link: https://bit.ly/3uzlZGd.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
We express our gratitude to the Universitas Indonesia which has funded this research through the scheme of Hibah Publikasi Terindeks Internasional (PUTI) Q2 No. NKB-676/UN2.RST/HKP.05.00/2022.