Water level control of small-scale recirculating aquaculture system with protein skimmer using fuzzy logic controller

ABSTRACT


INTRODUCTION
The recirculating aquaculture system (RAS) is a land-based aquaculture facility, either open-air or indoors, that minimizes water consumption by filtering, adapting, and reusing water. Compared to a traditional pond or open-water aquaculture, the water recirculation process in RAS allows for the control of culture conditions and the collection of wastes. In addition, land-based aquaculture avoids fish escapes and limits the transmission of diseases and parasites from outside [1].
RAS technology tends to be like a water treatment system where the recirculation process aims to control the quality of RAS water. Solid organic matter from fish waste and food waste directly becomes waste that needs to be eliminated because it is a source of increasing ammonia nitrogen (TAN), total suspended solids (TSS), total dissolved solids (TDS), and also has an impact on reducing dissolved oxygen (DO) [2]. TSS can be reduced by using mechanical filtration. TAN is lowered by using biological filtration through a nitrification-denitrification process. Meanwhile, TDS is lowered using a protein skimmer through a foam fractionation process. Proper aeration is necessary for the nitrification-denitrification and foam fractionation processes. Foam fractionation is the process of converting fine solids and protein residues in the form of fat into foam which is removed as FOAMATE [3]- [7]. RAS requires a water level control system so the fish tank does not experience water shortages or floods, disrupting the aquatic aquaculture ecosystem. Small-scale RAS can be analogous to a tank system so that controlling the water level requires a mathematical model approach to the tank system, which can be obtained empirically or experimentally [8]- [10] and the control strategy used [11]- [17].
The number and configuration of tanks in a tank system depend on the needs of the process to be modeled. The complexity and nonlinearity of a tank system model tend to increase with the number and configuration of tanks used. The selection of the control strategy and the type of controller used is crucial [9], [10], [13]- [15], [18]- [25]. In addition, this study involved a protein skimmer in RAS. The positioning of the protein skimmer in relation to the circulating water system in the RAS adds to the complexity of level control.
In this study, small-scale RAS is modeled using a 3-coupled tanks system approach with a tank configuration that follows the simplest RAS water recirculation process (fish tank, mechanic filter, biofilter). Clean water from the reservoir flows into the fish tank through a protein skimmer. This is intended so that clean water is completely free of TDS. In this case, the protein skimmer can be assumed as a supplier of clean water for the tank system. Meanwhile, the output water from the last tank (biofilter) is channeled into the reservoir. The purpose of control is to condition the water level in the protein skimmer at a certain level so that the foam fractionation process can work properly and control the water level in the biofilter tank in such a way that the fish tank does not run out of water or experience flooding.
This study applies the fuzzy logic controller (FLC) to control the water level in the protein skimmer and biofilter tanks by controlling the position of several valves where the placement positions of the said valves have been determined according to system requirements.

METHOD
This section explains the proposed research design and methods used. The proposed research design is the 3-coupled tank system and water level control. FLC is used as the method.

The 3-coupled tank system
The small-scale RAS built in this applied research is modeled using a 3-coupled tank system, as shown in Figure 1. The balance of water volume in the recirculating system is a problem in itself. Fish tank volume is usually larger than the volume of the water reservoir, with specific technical considerations. This difference needs to be anticipated so that the fish Tank does not experience water shortages, thereby disrupting the aquatic aquaculture ecosystem. By comparing the water level in the fish tank and reservoir and considering the water pump's capacity, the water volume balance in the recirculation system can be conditioned according to needs.
In this study, the reservoir is assumed to have free dimensions, so a balanced review based on water level is focused on the water level in tank 3. By conditioning the water level in tank 3 (bio-filter), it is hoped that tank 1 (fish tank) will not run out of water or experience flooding. Meanwhile, to ensure that the protein skimmer can carry out the foam fractionation process, the water level in the protein skimmer must be maintained at the specified liquid level. If the water level in the protein skimmer changes, it will disrupt the FOAMATE production process at the FOAMATE outlet.
Modeling a small-scale RAS with a protein skimmer in a 3-coupled tank system is carried out by observing the mass balance of water between tanks, with the primary supply source coming from the protein skimmer water outlet. From Figure 1, we can get (1).
(1) 1 , 2 and 3 is the constant valve position (0 ≤ ≤ 1). 1 and 2 are controlled using actuators. 3 controlled based on the change 1 and 2 ( 3 = 1 . 2 ). The pump water discharge specification is stated by . In protein skimmers, Bernoulli's principle applies as (2).  is the velocity of water flow at the protein skimmer inlet with a cross-sectional area 1 . . Because 1 " 1 . then it can be assumed ≈ 0. The outlet portion is at the bottom of the protein skimmer, so that can be assumed ℎ = 0. Then, (2) become (3).
The water discharge flowing at the outlet through cross-section can be expressed by (4), where = . √2. . The mass balance of the protein skimmer can be expressed by (5).
MATLAB Simulink Model from (5) is shown in Figure 2. The flow of water from Tank 1 to Tank 2 creates energy, where the total energy in Tank 1 equals the total energy in Tank 2. This can be expressed by Bernoulli's principle as (6).
Because Tank 1 and Tank 2 are open, the applied pressure is atmospheric pressure, so 1 = 2 .
The water discharge flowing into Tank 3 through cross section can be expressed by (11).
The mass balance in Tank 2 can be expressed by (12). 3 .
MATLAB Simulink Model from (12) is shown in Figure 5. In Tank 3, around the outlet pipe, it can be assumed that ℎ = 0 so that the applicable Bernoulli's principle is as (13).
The water rate in the outlet pipe is expressed by (14).
The mass balance in Tank 3 can be expressed by (15).
MATLAB Simulink model from (15) is shown in Figure 6, while the valve position variables, 1 and 2 are modeled as shown in Figure 7.

Water level control
In this study, 1 is the variable valve position used to adjust the water level in the protein skimmer (ℎ 1 ). 2 is used to adjust the water level in tank 3 (ℎ 4 ), while 3 is used to adjust the multi-valve manually. The multi-valve regulates the water flow transition from the protein skimmer to the tank system. The water level in the protein skimmer is expected to reach ℎ 1 = 50 , while the water level in tank 3, ℎ 4 = 40 . The specifications of small-scale RAS with a protein skimmer are shown in Table 1.

Fuzzy logic controller
FLC is a fuzzy logic-based controller that implements the fuzzy inference system (FIS) in its reasoning. FIS consists of four parts: fuzzification, inference engine, knowledge-based, and defuzzification. Fuzzification is the process of transforming crisp values into fuzzy values using fuzzy sets for each of the variables involved. Fuzzy sets are constructed using membership functions (triangular, trapezoidal membership functions, etc.) that represent the linguistic variables used. The linguistic variable is a representation of human reasoning over many values within a specific range. An illustration of the fuzzification process is shown in Figure 8 An inference engine is a reasoning process based on the logic of human linguistic knowledge (knowledge-based) packaged into a rule-based set. The inference engine applies fuzzy operators, implications, and aggregations. All three use the AND/min or OR/max method, as needed. If the input is more than one, then the fuzzy operator needs to be applied to each rule-based line. The implication is the transformation of the result of using the IF (input) THEN (output) rules to each rule line into a set of fuzzy outputs. The implication method can be AND/min or OR/max. As the illustration shown in Figure 9. Aggregation is the process of combining all output fuzzy sets into a single fuzzy set. Typically, aggregation uses the AND/max method. Finally, defuzzification is the process of transforming a single fuzzy set into one crisp value as the FIS output. Usually using the centroid method. This is illustrated in Figure 10.

RESULTS AND DISCUSSION
These results are obtained from the simulations carried out using the MATLAB/Simulink software. The results are divided into five parts, single-input FLC, single-input FLC with setpoint, gained single-input FLC with setpoint, tuned single-input FLC with setpoint, and setpoint change experiment.

Single-input FLC
Single-input FLC is an FLC that utilizes water level output data at any time as input. While the FLC output is a constant valve position (0 ≤ ≤ 1). Linguistic variables for the fuzzy input set (Level) are VL  where is the normalizer value. FLC with one input was built using FIS MATLAB with some design characteristics: i) FIS: Mamdani, ii) Implication method (AND): min, iii) aggregation method: max, and iv) defuzzification method: centroid. MATLAB Simulink model for water level control small-scale RAS with protein skimmer using single-input FLC is shown in Figures 12 and 13, with output response water level in the protein skimmer as in Figure 14

Single-input FLC with setpoint
If a setpoint is included for each water level (ℎ 1 dan ℎ 4 ) then it is necessary to consider the presence of an error level, which is expressed by: ℎ = ℎ − ℎ . With ℎ acting as a normalizer such that _ = ℎ /ℎ , then _ will be within the value range {-1 … 1}. The set of fuzzy inputs (error level) used is shown in Figure 15   Output response ℎ 4 ( = 9.77) Figure 18. Output response using FLC one input with setpoint

Gained single-input FLC with setpoint
To reduce the FLC gain is used, which is placed on the output side of the FLC, as shown in Figure 19. In this case, the controlled signal is expressed as (17).
FLC Protein Skimmer FLC Tank System Figure 19. MATLAB Simulink model for sub-system gained single-input FLC setpoint The system response output is shown in Figure 20. By setting 1 = 1.16 and 2 = 1.6 a protein skimmer water level steady state value can be achieved, ℎ 1 = 50.03 with = 77. 25 , as can be seen in Figure 20

Tuned single-input FLC with setpoint
Next, is reduced by tuning the input and output fuzzy sets (without FLC gain). Tuning the fuzzy input set is done by changing the range of values for each linguistic variable, as shown in Table 2. The form of the membership function is the same as in Figure 15. Meanwhile, the output fuzzy set tuning is shown in Table 3 and Figure 21. The system response output is shown in Figure 22.  . The appearance of overshoot in the response output ℎ 1 is still acceptable because it is still within the ±2% area of the final value. A summary of the experimental results of dynamical system models using several types of singleinput FLC with setpoints is shown in Table 4. From the table, it can be seen that the tuned single-input FLC has the best average output response characteristics.

Setpoint change experiment
Furthermore, both types of FLC (gained and tuned FLC) controllers were tested with changing setpoints. This test is intended to simulate the condition of the small-scale RAS water level, which under certain conditions, may change because the mechanical filter material begins to saturate with TSS. The test results, which are shown in Figure 23, show that tuned FLC can follow setpoint changes well, and vice versa for gained FLC.

CONCLUSION
The small-scale RAS mathematical modeling in this study is approached by analogy with a 3coupled tank system (fish tank, mechanic filter, biofilter). System dynamic models become complex when protein skimmers are involved. In order for the foam fractionation process in the protein skimmer to function properly, the water level in the protein skimmer needs to be conditioned in such a way as not to change. In this study, clean water from the reservoir flowed into the fish tank via a protein skimmer to ensure that TDS was completely removed. This means protein skimmer is a source of fish tank supply. Because the biofilter tank is the last tank where the water outlet goes to the reservoir, the water level of the biofilter tank needs to be controlled in such a way that the fish tank does not lack water or otherwise flooding occurs. Control of the water level at these two points is carried out simultaneously through the configuration of several predefined valves where the opening/closing position of the valves is controlled using FLC. This study uses a singleinput FLC with a setpoint designed in two modes: gained and tuned FLC. The study results show that the tuned single-input FLC has the best average output response characteristics. Tests were also carried out by providing a changing setpoint, where the results showed that the tuned FLC could follow the setpoint changes well and vice versa for the gained FLC.