Early Fault Detection Experimental System
Albert Setiawan and Jennifer Mathieu

The greenhouse
The aerial environment of the greenhouse is controlled by a computer. The environment variables controlled are light integral and temperature. The environmental variables monitored include relative humidity, temperature, carbon dioxide, and light intensity. The temperature set-points were 19 ° C during the night and 24 ° C during the day. These set-points were achieved by ± 0.5 ° C. The light integral set-point was 16 mol m^-2 day^-1 of PAR, which was achieved even during the wintertime by using supplemental lighting.

 
Figure 1. Greenhouse under supplemental lighting.

Tanks
The hydroponic system used is called the deep trough system. In this system, the tank is filled with nutrient solution and the plants are grown on Styrofoam with holes, which floats on the surface. This system mimics the production system used in the Controlled Environment Agriculture demonstration module. However, instead of harvesting and transplanting everyday, the interval is every two days due to space limitation. Three stainless steel tanks with dimension 2 by 4 feet were used in this project. The stainless steel tanks minimize corrosion and leaching of minerals into the nutrient solution. One of the tanks was the control and the other two were used for fault treatment.

 
Figure 2. Ponds with lettuce (ages 12 – 28 days).

Circulation System
In a deep trough system, aeration is necessary since the water surface is completely covered by Styrofoam, which minimizes evaporation and discourages alga growth. Pure oxygen is injected into the system to ensure that enough oxygen is supplied to the roots, without which the level could drop from 6.5 to 1.5 in one day. A distribution system was needed to evenly distribute the oxygen and nutrients in the system and to insure the sensors have adequate flow to work properly. A pump was used to draw the solution through a filter and then pump it through the PVC distribution system along the perimeter of the bottom of the tanks. The pipe has small holes, which encourages mixing. This is important especially because the pipe is used for acid and base injection to maintain pH.

 
Figure 3. Piping and filter of distribution system inside the tank

Seedling and spacing procedures
A continuous lettuce production system was selected to achieve nearly constant conditions in the experimental tanks. Every two days seeds were sown, 12-day old plants were transplanted into the system, the plants in the system were re-spaced, and 28-day old plants were harvested. In an effort to keep the system as clean as possible, rockwool with a dab of peatlite on top was selected over a cell of peatlite. The rockwool fibers do end up in the nutrient solution, but at a greatly reduced rate compared to the peatlite. This improvement greatly affects the life of the nitrate analyzer. The rockwool is cut to fit a 1" cube black cell. Peatlite is then inserted in the factory-drilled hole and gently patted down to get a consistent firmness. The lettuce seeds are placed with tweezers on the peatlite and each one is pressed to insure full contact with the peatlite. They are then sprayed with reverse osmosis water and covered with a humidity cover for 2 days. On day 7, the plants were selected based on size. When the seedlings are 12 days old, they are transplanted into the experimental system. Spacers (¾") are used to give the plants enough room to grow and are added incrementally. The following Table shows how the spacers are added. Finally, the 28-day plants are harvested from the system.

Table1. Spacing for continuous lettuce production.

Meters and signal conditioner
The monitoring and control system used consists of meters with their corresponding sensors. The meters are connected to a signal conditioning unit from National Instrument and then to a dedicated computer. The control and monitoring program was written in the G programming language of LabVIEW 5.0 also from National Instrument. The sensors could, in fact, be connected directly to the conditioning unit. However, some of the sensors have very high impedance and need a special signal conditioning module to handle them. Since the meter price is usually cheaper than this special signal conditioning module, meters were selected. In addition, the meters can be calibrated directly in case the computer is broken down.

Figure 4. Meters used to monitor one tank. From left to right: Scale, pH, EC, DO.

Computer and program
A dedicated computer is used to control and monitor variables in the nutrient solution which include temperature, electrical conductivity, pH, dissolved oxygen, nitrate concentration and nutrient solution volume. The program used is LabVIEW 5.0 from National Instrument, which is a leading industrial programming language for control and instrumentation. The program controls and monitors the nutrient solution every 10 seconds and logs the data every 5 minutes.

 
Figure 5. Display of control and monitoring program.

Sensors

pH
pH is considered the easiest variable to measure in solution. However, we found little agreement in sensors advertised to have an accuracy of ± 0.01 pH units. The differences in measurements were usually ± 0.1 pH units. The first sensor from Omega was not fluoride resistance and we were advised to use a fluoride resistant sensor after one of the membranes shattered in the solution. However, we have no reason to believe that we have high fluoride levels in our nutrient solution. Using the same meter, the standard sensor and the fluoride resistance was found to have a 0.1 difference in their readings. The set-point for pH is 5.8.

EC
Electrical conductivity is usually measured with electrode type sensors. Although the accuracy is very good, they need frequent maintenance since deposits built up on the electrodes. Since our system is used continuously and every maintenance means disturbance to the plants, a different kind of sensor was selected. This sensor is based on different magnetic fields caused by different electrical conductivity of the solution. It was found that the accuracy was relatively good at ± 25 m S/cm and it required very little maintenance. The set-point for EC is 1200 m S/cm.

DO
At first, a sensor and meter from Omega was used in the system. The sensor size was small compared with others thus very attractive to be used in a small tank. The meter was also waterproof. However, it turned out to be very inaccurate and unreliable and the output of the meter was unreadable after a few months. The sensor’s impedance was also very high at about 16 MW . A new type of sensor was used which was 3 times bigger in size but had very low impedance 500kW , which could even be connected to the signal conditioning module directly. The accuracy and reliability was very good and only needed cleaning after it had not been used for months. The set-point for DO is between 6.5 to 7 mg l^-1.

A stationary rack was designed to hold the sensors in the correct orientation in the flow pattern to improve the accuracy of the pH, EC, and DO measurements.

 
Figure 6. Sensor rack in the tank: From top to bottom: pH, DO, EC.

Nitrate
Monitoring nitrate concentration in the nutrient solution is desirable since nitrate uptake mirrors plant growth. This also makes it a good indication for plant stress. There are very few ion selective electrodes available for continuous monitoring. The only system found for continuous operation was from the Hach Company, Inc. The incorporation of the analyzer in the system turned out to be very difficult. The analyzer was designed to be used for drinking water, although the company said some customers use it for wastewater measurement. Solids and algae in the solution cause a lot of problems. At first, a mixture of peatmoss and vermiculite (peatlite) was used for the seedling production. To reduce the solids in the nutrient solution, the peatlite was replaced by rockwool. Frequent maintenance is needed to insure the analyzer has a clean sample and the electrode calibration is functioning properly.

Evapotranspiration
Evapotranspiration was measured by placing the whole tank on a scale. The evapotranspiration rate can be calculated from how much the weight changes over a given period of time. The accuracy of the scale used is ± 50 g and it is still accurate after 6 months of usage. The scale was connected to the computer directly using RS232 serial connection. The set-point of the tank weight was 127 kg.

Fault Detection in hydroponics plant production system using fuzzy logic algorithm
Today, real time fault detection is becoming an important technology. In the past it was used in critical systems such as nuclear power plants, and other applications where system stability is absolutly vital. With current advances of computer technology, real time fault detection is becoming feasible due to the massive computing power of simple PC's. The computer is able to perform controlling duties, while running a supervisory program to detect faults in the system.

In a CEA hydroponic plant production system, most environmental variables are controlled automatically (via computerized climate control), therefore real time fault detection is becoming more important as nobody is supervising the greenhouse all the time. System faults can range from a broken circulating pump that is easily detected, to the drifting of sensor that is a much harder to detect.

There has been some research conducted that has addressed fault detection of this type. They are divided into two catagories: hard fault and soft fault. A hard fault is quite obvious, and is simply when an instrument or device stops working. A soft fault occurs when equipment is working, but with some error associated with it.

In a hydroponic plant production system, the plants interact closely with its environment. In the tank filled with hydroponic nutrient solution, variables such as pH, Electrical Conductivity, temperature, dissolved oxygen, and water level are controlled using a computer. The computer system continuously collects the data relating how the plant is interacting with its environment. The preliminary experiment showed a normal cycle of nutrient uptake and evapotranspiration in lettuce. For a plant under stress it was hypothesized that these variables would be different from the normal non-stressed pattern. This correlation is an early indicator of a plant stress, therefore the problem can be caught and remedied before visible signs of stress. Early fault detection makes it possible to fix a problem before the effect of the stress/fault in the system becomes irreversible. In this case, using fuzzy logic to detect fault in the plant itself could be beneficial since the biological system could not precisely be modeled mathematically. Fuzzy logic uses membership functions to model the impreciseness of each model and then uses rules applied to these variables to detect plant stress. The proposed thesis will analyze hourly data and will look at the variables in real time. It will then be compared with historical data to determine if the system is in normal operating condition.