Students: Elon Hendriksen, Tiemen Markerink, Guus Fidgor, Arian Duine. (2017)
Original Situation
Our client, Airborne, has the goal to make carbon products and production more affordable. This mostly by developing automation and robotization based solutions to replace the current labor intensive processes.
Currently, airborne uses coordinate information from xy-cutting machine placed earlier in the production line combined with an encoder, thereby keeping track of ply movement along the conveyor belt. After a ply is within the workspace of the robot, it is located and picked with the right amount of suction cups based on the above information flow.
Problem statement
The above described situation uses a passive system and does not monitor unwanted ply movement which can occur, for example, as a result of sticking plies during pick up or conveyor belt slippage.
Therefore, the accuracy of picking up plies is obstructed, resulting in potential product damage. To obviate this, an active system using vision software was preferred by our client. With regards, the following objective was set;
“Within 8 weeks a more reliable recognition and pick-and-place method for cut out pre-impregnated plies must be researched, developed and demonstrated. This with the usage of Vision, a process display and fitting the current production setup of Airborne.”
Goal of the project
Like described above, the goal of this project is to create a system that is able to adapt to slight un-expected and unwanted movements in the plies that occur during transport or pickup.
Requirements
Demands
- All the preformed tasks of the end product need to happen within the reach of the used robot arm.
- The robot is able to determine locations of pre-cut plies with the help of a vision camera.
- The robot is able to identify the shape/code of the plies.
- The robot activates the right suction cups depending on the shape of the ply.
- The robot will use an ‘end of arm tool’ that utilizes suction cups to pick up the cut-out parts and is able to do so without damaging the sheets.
- The delivered system is able to place the cut-out plies on a stack in an ordered fashion.
Wishes
- Make the robot work as efficient as possible.
- Try and make the proof of concept look as similar as possible to the situation at ‘Airborne’
- The final system is delivered as soon as possible.
Our solution
At the end of our 9 week project, the complete active system is working as demanded, easily passing the set client criteria.
The In-Sight Program
The cognex program is working with a binarize filter which adjusts the threshold automatically due to the light intensity. After that, an erode function will be applied. This will make one blob of the code with an area between 13000 and 16000 units. The blobs have a requirement of a shape that has the width around 270 and a height around 70. With this requirement the program will know whether this blob is an actual code or not. When it is not, the blob will be of no further use.
When the blob is a code, the “T” recognition function will search for a “T” in the area of the blob. This function will determine the location and orientation of the code. With this location and orientation, a OCR Max function is being put on the code to recognize the actual code. The recognized code is compared to the code in the database. The database is a simulation of the received information from the nesting machine of the actual Airborne system. This database contains information about the offset of the “T” to the robot pick up point. The suction cup selection is also in the database, which simulates a program at airborne that calculates what suction cups it should use.
When it has found a positive result, the final coordinates will be calculated with the found coordinates and the offset coordinates in a certain angle, which is determined when recognizing the “T”
Next a data string will be set up with these coordinates and the suction cup selection that belongs to the code that has been found.
The system
The Cognex camera makes four pictures per second and analyses them for codes. When a code is recognized, a data string will be put together.
The robot gets a data string which tells that a code is found. When the robot receives the data string, it will send a signal to stop the conveyor belt. After the conveyor belt is being stopped, another picture is taken to get the new location of the ply, Because the original picture was taken before stopping
the conveyor belt.
With the information of the data string, the robot moves above the ply and rotates its end of arm tool in the correct way. Then it moves down to the ply and selects the correct suction cups based on the code on the ply that was recognized before. The robot goes up very slowly to ensure the ply is taken with ease. After that the robot can accelerate to a higher speed.
Next, the robot puts away the ply on a stack next to the conveyor and resets to its home position while the conveyor is started again and everything starts again until the conveyor is empty.
The robot of concern is able to determine prepreg ply locations with the help of vision, which iden
tifies the ply codes and gives the coordinate, suction cup and offset data within 250 milliseco
nds. Where after, the robot accurately picks and
places the identified ply into neat stack next to the conveyor belt. In addition to the active system, unwanted movement due to sticking plies is limited with a complaint retaining
mechanism placed on the conveyor belts sides. All combined, results in reliable, fast and within 3 millimeter accurate robot setup.
Major Decisions
The way of ply recognition with vision was an important decision made during this project. With options as shape recognition, simple blob recognition or Optical Character Recognition (OCR), eventually a combination of all was made due to accuracy and progressing time optimization. Along with smart filter and region applications of the above functions, thereby limiting image distortions.
Another major decision was the choice of the robot. The ABB Delta robot was chosen for this project. This Robot is new to SMR. That means that the team had to find their own way to get the robot working, instead of using the know knowledge when another robot had been chosen. The End Of Arm Tool was also limited to 1 Kg due to the capacity of the delta robot.
Conclusion
At the end of our 9 week period the complete system was working properly. When looking back at the demands made in the first weeks of the project we can easily say that the robot passes on all of them.
- The proof of concept is finished at the end of the 9-week period.
- All the preformed tasks of the end product need to happen within the reach of the used robot arm.
- The robot is able to determine locations of pre-cut plies with the help of a vision camera.
- The robot is able to identify the shape/code of the plies.
- The robot activates the right suction cups depending on the shape of the ply.
- The robot will use an ‘end of arm tool’ that utilizes suction cups to pick up the cut-out parts and is able to do so without damaging the sheets.
- The delivered system is able to place the cut-out plies on a stack in an ordered fashion.
To conclude we can all say we are very pleased with the end result and we are sure that our client can implement our gained knowledge usefully. Our gained knowledge will be described in a small report for Airborne.