Team members: Rowan Overdevest, Guus Paris, Bram van Riessen (2016)
Automating the sorting of orchids
Automating the accurate identification of topological features of organic products is indeed a difficult task. But no task is too hard for the Roborchid team.
Hazeu Orchids is an innovative agricultural company, always looking for improvements in their greenhouses where they grow 120.000 orchids every week. There are a several moments in the orchid growing process where the orchids have to be checked and sorted. During a check numerous features of the plant are determined depending on the growing phase of the orchid. They check for the size of the leaves, plant-height, the amount of flowers and the amount of spikes. These features determine the quality of the product and the final selling price of the orchids.
One of the check-ups is focused on determining if a plant is large enough to start growing a spike and flowers. This process is currently executed by three employees, who are dividing the plants in two groups; big plants and small plants. After dividing, the plants are placed on the correct conveyor belt.
The assignment is to automate this process, the sorting of orchids on size of the plants. Roborchid is a project team that consists out of three ambitious mechanical engineers with one common goal. To build a system consisting of a robotic arm and camera vision combination that sorts the orchids on size.
In short, our goal is Automating the sorting of orchids.
Future situation, the designed automated system
The complete designed system consist of four subsystems:
- the input subsystem
- the transport subsystem
- the inspection subsystem
- the output subsystem
The input subsystem translates the input so that the rest of the system will function. The transport subsystem controls the flow of the orchids form the input threw the entire system towards the output. The inspection subsystem controls the decision making and controls through which output the plant will leave the system. The output subsystem is the physical parts that are controlled by the inspection subsystem.
Orchid sorting system:
The goal of the input subsystem is simply to accurately determine the position of the orchids when they enter the system. This is done so the plants have a fixed position that the sorting system can find. This is achieved by creating an additional grid on the tables of the current situation in which the plants will be fixed on the tables.
After the orchids entered the system. The transport subsystem will take care of the transportation of the orchids throughout the system. The transport subsystem consists of the wheels that control the position of the tables, the Universal Robot that picks up and places the plants from the table on the conveyer belts, the conveyor belts, the rotation strip, a pneumatic actuator and a Siemens PLC that controls all mechanical parts.
The inspection subsystem consists of a Cognex camera vision system that checks the rotating plants for various parameters and gives a signal to the Siemens PLC that controls the ouput.
The output subsystem consists of a linear actuator that controls a physical switch that is located at the end of the conveyor belt. The PLC combines a signal of the linear actuator from the transport subsystem with a signal from the Inspection subsystem it’s Cognex camera to either open or close.
Major project decisions
The final form of the designed automated decision was formed based on a couple of major project decisions.
The biggest obstacle of this project was deciding how the inspection subsystem would work. Deciding if a plant is either big or small is a subjective feature and depends on the interpretation of the sorter. So the first thing to do was calibrate our definition of what a big and small plant is to be the same as the responsible person of the Hazeu sorting department.
To decide how our system would sort plants we made a very large spreadsheet with data from a large variety of plant features. We measured among other features the weight, visible length, stem length, leaf lengths, width, colour and leaf shape. And we considered numerous system measurement methods among which, kinect 3d scanner, using a 3d camera, making pictures from above or even holding the plant upside down in water to measure the water level rise.
Finally we found a correlation between the location tip of the leaves and the size of the plant when a picture from the side of the orchid is made. This method increased our correct sorting-rate to 95,5%, from 80% with other methods.
Another major project decisions was about the method of picking up the plants from the tables. How the first part of the transport subsystem would work. Because the plants only have to make linear movements over two-axis to be placed from the table on the conveyer belt we have considered building a simple two-axis gantry robot, that could be cheaper than a anthropomorphic robot-arm. Finally we concluded that building such a machine to professional standards would not be cost effective.
The last major project decision was how the input subsystem would work. Preferably we would incorporate a second camera that would communicate the location of the plants to the system. This would though not fix the problem of plants falling down or plants sliding on the tables into positions the transport subsystem would not be able to pick up the plants.
So we decided to incorporate a method that would keep all the plants on a table in a fixed position. When in combination the fixed position of the orchids the table is moved by the transport subsystem, the location and partially orientation of all possible plant are in the system.
To test the designed system a proof of concept has been built and tested with over 300 different plants. During testing the system has been calibrated and evolved. Every subsystem and the total system have been tested.
part from concluding that the system needed it’s necessary changes, the most important conclusion of the test is that the error-rate of the sorting process will improve with the automated designed system compared to the current system. During testing the automated system was able to achieve an error-rate of 4,5%, which is better compared to the measured 7% error-rate of the current system.
Conclusion (Business case)
An accurate prediction of the designed automated system it’s speed, sorting error-rate and costs are achieved by making calculations and testing the proof of concept. When we compare these numbers with the current system we gain the following results:
Reviewing the results of the project and seeing that we achieved all set goals and demands from the brief of requirements we can happily conclude that the project is a succes!