literature review

Building a model to control a process is extremely important for any industry because industries have to meet the effluent requirements of the plant. The effluent requirements could be determined by governmental institutions, such as European Union, to protect the environment, or they could be determined by the customers who buy the effluent product, such as refineries products. The effluent standards lead to increase the operational costs and economical paneities. Therefore, designing a proper controller that represents the process accurately to maximize the profit and avoid penalties is essential.

However, not all controllers can be designed easily because there are a lot of processes behave in a non-linear manner. Also, the influent may experience a remarkable perturbation in flow, load, and composition. Consequently, this work is important in academic point of view because it teaches a new technique to solve problems and design a controller for wastewater treatment system. This work teaches process control engineers a way to design a controller for abnormal conditions.

The controller used in wastewater treatment is Model Predictive Control, or MPC, which is a computer control algorithm that predicts the future response of a plant by employing an explicit process model. It yields good results for both linear and non-linear predictive control technologies. Therefore, model predictive control is a good representation for the oxygen control of wastewater treatment plants.

The modelling of the biological reactions used to simulate the biological reactions in aerobic and anoxic reactor is Activated Sludge Model 1, or ASM1, and double-exponential settling velocity function is used in the second settler of waste water treatment plant to model the clarification and thickening processes. The modelling of the secondary clarifier is flux-theory which is one-dimensional model. This model assumes uniform horizontal velocities so the horizontal gradient in concentration is negligible, and negligible biological reactions. Therefore, only the vertical dimension processes are modelled.

The model of the aeration process has to be accurate representation of the process because aeration process is very critical for the entire activated sludge process. Microorganisms need enough oxygen so that there is enough electron receptor capacity for their metabolism process. The process of oxygen transferring from air bubbles to microorganism cells is complicated. Therefore, the slowest process, which is convection of mass transfer within the air bubble to the gas liquid border surface, was chosen as determining factor for the whole process. A dissolved oxygen mass balance model was used around a complete stirred tank reactor which uses oxygen mass transfer coefficient as manipulated variable.

To control the dissolved oxygen concentration at a certain level, the following process model is used. First, the concentration of oxygen in the reactor is measured by an ideal sensor. Then, the concentration value of oxygen is handled by the control method to calculate the oxygen mass transfer coefficient. Then, the oxygen mass transfer coefficient is corrected to match the corresponding operational temperature. Finally, the oxygen concentration level in the biological reactor is changed by applying the oxygen mass transfer coefficient. As a result, the volume of the air blown by the diffusors and the operational cost for the aeration can be calculated.

Model Predictive Control is one of the classes of algorithms that optimize the future behavior of a plant by calculating a sequence of manipulated variable adjustments. The controller design model linearizes the aeration process in ASM1 model at a steady state operation to build a state model of the waste water treatment plant. The manipulated variable in the controller is the oxygen mass transfer coefficient and the output is the concentration of the dissolved oxygen. The sensor of the oxygen is ideal without time delay and no consideration of noise is taken. The aeration process has a second order model which was proved to be sufficient representation of the real aeration process. The performance assessment of the model is performed using integral of absolute error, or IAE and integral of square error, or ISE.

Model Predictive Control has difficult concepts that beginners to process control may not be able to understand. The following is a list of questions about MPC. How is the first input in the optimal sequence of MPC is calculated? How does the constants m and p are being optimized to minimize the quadratic objective? What are the components of y and u that could be penalize by the weighting matrices in this case? How would the other tuning parameters, like control and prediction horizon and weight matrices, affect the performance of MPC controller? Why is the sampling time in control of the simulation benchmark has as significant effect on the performance of the controller?

Model Predictive Control of the dissolved oxygen concentration shows successful results in an aerobic basin of a pre-denitrification process with influent disturbances and in an alternating activated sludge process. According to Copp, the benchmark simulation results and Model Predictive Control results for control strategy of activated sludge plant agree with each other and give similar results (Copp, 2020). Predictive model control has many advantages. For instance, it can follow the rapidly changing dissolved oxygen, or control variable, setpoint. Also, manipulating the parameters of the controller can decrease the error between the output and the set point which yield better results. Model Predictive Control can also solve problems for linear and non-linear systems without changing the controller design. Moreover, performing step test in Model Predictive Control is sufficient to build a model and obtain its parameters.

Even though Model Predictive Control has many advantages and gives good results, it still has some drawbacks. For instance, a lot of assumptions have been made to build the process models which might yield inaccurate model to represent the real process. For example, in the aeration process model, the mass transfer of oxygen within the liquid phase to the microbial flocs model was neglected because it is faster than the other processes happening in the aeration process. What is more, some of the input variables are separated to make the process model simple, which might affect the final results of the controller. For example, the only manipulated variable considered is the oxygen mass transfer coefficient, and all other inputs to the reactor are separated and considered as unmeasured disturbances. Also, some of the parameters of the controller are tuned by using trial-and-error method which might be inaccurate and time-consuming way to obtain data.

Some possible improvement can be done to the controller to make its performance better and more accurate, even though it might be more difficult to obtain. One suggestion is to consider the biological reaction in the model of secondary clarifier since there will be sufficient oxygen concentration in the fluid. Another suggestion is to consider more inputs to the reactors of waste water treatment, not just the oxygen mass transfer coefficient. For example, the microorganisms are affected by temperature, PH, and many other factors that should be considered in modeling the process.

In conclusion, Model Predictive Control is an accurate strategy to control a dissolved oxygen concentration. It was tested in two simulated case studies, one is to control the dissolved oxygen concentration in aerobic basin of a pre-denitrification process with influent disturbances, and another is in alternating activated sludge process. Both studies show successful results of the controller. This work is important for industries to meet their effluent specifications; what is more, it shows students and process control engineers a strategy to control a non-linear process. The controller used is Model Predictive Control and different models were built for different units of the plant. The model has some drawbacks and limitations, but it still gives reliable results.

References

Copp, J. B. (2002). The COST simulation benchmark: Description and simulator manual (COST Action 624 & COST Action 682). Luxembourg: Office for Official Publications of the European Union.