Machine Vision for Pellet Sorting in Recycled Plastics

Recycling plastics has become commonplace virtually throughout the world. Challenges to the reuse of the recycled plastics for the equivalent of the original application for the specific plastic type (e.g., as a container) include making sure the batch is not contaminated by both foreign material and foreign colors. Metal detectors can detect metallic foreign material. Spectroscopic techniques are used to eliminate foreign non-metallic materials. Materials of foreign colors are eliminated using machine vision techniques. Color contamination can result in a decrease in the value of a batch and even scrapping the entire batch.

Typically the recycled plastic becomes a sea of pellets. In some ways the machine vision systems used to separate foreign colored pellets resembles a web scanner or a system that inspects products produced in a continuous sheet. Another analogy is to the machine vision systems used in sorting food products. In all cases, product is delivered in continuous motion to the machine vision-based scanner system.

It seems that sorting of recycled pellets is of more concern in Europe than in the US as most companies supplying such systems appear to be European-based. Two companies provided input for this article.

• Joerg Schunicht – Sales & Project Manager – CommoDaS
• Johannes Giet – Head, Surface Inspection R&D – ISRA Vision Systems

1. How would you describe your machine vision-based plastic pellet sorters? What are the specific products you offer that address these applications?
[Joerg Schunicht – CommoDas] Our MikroSort® separation equipment programme covers systems for colour removal as well as for foreign material rejection. For instrumentation of our equipment true colour cameras, near infrared (NIR), x-ray and eddy current sensors are available stand alone or in combination. The design of our machine vision hardware and software platform is based on more than 12 years experience in sorting of bulk materials. The machine vision system is important as it yields an outstanding sorting efficiency. In addition, the right feed presentation and the high-speed compressed air separation provides advantages by eliminating the removal of good product.

[Johannes Giet – ISRA Vision Systems] ISRA’s new color-based Sorting System (SORTexpert  ) is the result of 20 years of experience in machine vision. Its platform uses the latest state-of-the-art high-speed processing, high-resolution digital color cameras and combines sophisticated sorting techniques and functions implemented on FPGAs. Due to the outstanding bandwidth of our image processing boards and the real time working color-processing functions the system is able to separate bad or foreign objects at a tremendous throughput. For complex requirements, appropriate laser scanners or multi¬sensor systems are available.

2. What specifically differentiates your machine vision products that address plastic pellet sorting?
[Johannes] To meet the demands of an ever-changing market we have designed into our system a strong, flexible hardware platform based on FPGAs that is easy to expand and modify. This means we can respond to the ever-increasing high performance demanded in the marketplace today and in the future. Typical color systems are only capable of measuring the size, position and the average color.  SORTexpert measures and determines the size, position, color, shape, structure, texture, i.e., more than 10 features of each object (100.000 Obj/sec).

Where a traditional quality sorting classifies based on rules, SORTexpert can be trained by the user by simple “showing” examples to the system (simple “click and learn”). Automatic calibration tools deliver a consistent sorting efficiency and performance.

[Joerg] Our machine vision solution is an in-house development specialized to the needs of image based sorting of bulk materials. The processing pipeline can handle three imaging sources in parallel, correlate the data of sensors with different resolution taken at different positions and process material features in one classification model. For small grain sizes such as plastic pellets, a load proof processing path is available making sure there is no performance limitation caused by the image processing system.

3. Can you briefly review the underlying principles associated with your machine vision-based products? Specific hardware arrangement and software?
[Joerg] Our sensors are supported by sophisticated image pre-processing performing the specific pixel-based operations such as calibration, transformation, classification and filtering. On the resulting feature classes a segmentation or a filter processing can be performed. In some applications both methods are applied in parallel order yielding speed advantages. This underlying principle is applied to all of our sensing systems that can be true colour, NIR or x-ray imaging or a combination.

[Johannes] Depending on the task, multiple color (2K, 4K) or B/W linescan (up to 6K, 8K) cameras with the most modern technology and the highest possi¬ble resolutions are used to capture images for classification and sorting. The cameras are positioned above a belt (depending on the products different views are sometimes necessary) and capture the images of the products, when they pass. These cameras detect discolorations or deviations on or in-between the product stream.

Data processing is carried out by ISRA's own stan¬dard processors with a data input of 160 Mbytes/sec. With these processors, complex processing steps such as filtering, color space conversion (HSI), feature extractions are executed in video real-time at an internal data rate of 800 Mbytes/sec. Using this signal processing, the sorted products are already inspected for defects and clas¬sified during the capturing process. Information that has been reduced to the fundamentals in this way facilitates a high sorting quality at highest throughput.

4. What are the critical performance parameters of a machine vision-based plastic pellet sorter that a customer should understand? Throughput? Performance in terms of percent sort effectiveness? Other parameters?
[Johannes] One critical performance parameter is of course the throughput with respect to sort effectiveness. Nevertheless the CD and MD resolution of the total system along with the performance of the classifier has an important impact.

[Joerg] The limiting factors relate more to the mechanical behavior of the feed presentation and the image capturing than to the data processing. The overall performance depends on a perfect integration of the imaging solution in the equipment design. This is why CommoDaS not only builds the machine vision but also all the surrounding material handling.

5. Are there specific application issues that a customer should understand to optimize system performance? Particle sizes? Shapes? Presentation state? Sorting mechanism impact on sorting good pellets? Etc.
[Johannes] As mentioned, the particle size and shape in combination with the lighting concept is important. For special applications different views are imperative.

[Joerg] The narrower the grain size range the better the separation results are. That means the customer should remove oversize and undersize material by efficient screening facilities. Furthermore the feed rate should be controlled in order to run the sorting system at the most efficient capacity and avoid overload conditions.

6. Are there specific hardware parameters that a customer should understand? Optics issues? Camera issues? Lighting issues? Software issues? Etc.?
[Johannes] Lighting and optics are 70% of most machine vision applications. All machine vision systems have one issue in common, this is appropriate lighting. For our systems different standard lightings are available. State-of-the-art is LED lighting due to the lifetime and capability to control the intensity.  The sorter has different modules/functions to control the lighting i.e., adaptive lighting control. This module is extremely significant to get consistent results.

[Joerg] Based on the customer’s requirements CommoDaS defines the equipment configuration including feed presentation, illumination, optics and sensor arrangement. 

7. What are the skills required to set up a job and operate a machine vision-based plastic pellet sorter?
[Joerg] Once our application engineers have set up the equipment and trained personnel on the application, a person with normal technical understanding and a basic training can run the equipment.

[Johannes] In general the skills required are highly dependent on the system software and the settings to be made. ISRA doesn’t sell “project tailored” machine vision systems. Based on our experience developing machine vision products we have a sophisticated operator guidance e.g., to train the classifier (click and learn).

8. How do you support your products – training, documentation, warranty, post installation service, software revisions? Are these free or is there a fee?
[Johannes] In general, training is provided. Each system is shipped with a full set of manuals. The documentation is available either online or hardcopy or both. All systems come with a standard warranty. Service contracts are available. As one of the leaders in machine and surface vision ISRA has its own customer support centre (24/7 telephone hotline, tele-service, training, spare parts service …).

[Joerg] The equipment set up and the basic training is included in our scope of delivery. For maintanence and assistance we offer service contracts with guaranteed response times and telephone support. Service cost savings using our remote maintenance concept via Internet have been remarkable.

9. Where do you see breakthroughs coming in the specific infrastructure technologies (hardware and software) that are the basis of the machine vision technology (all hardware and software) you use in the near future – next three years?
[Joerg] The machine vision system development takes advantage of high speed integrated circuits available in the market. So processing power is no longer a bottleneck. In the future new sensor technology will drive applications measuring the material features more direct and more reliably with high-speed and low noise.

[Johannes] Machine vision technologies have benefited from advances in higher resolution cameras, special color cameras, in lighting leading to LED lamps and in PC and FPGA technology. At the same time machine vison is becoming less expensive. Our opionion is in the next years we will see more and more smart cameras/systems and hopefully the first systems on chip.

10. What specific performance improvements in the systems you sell are anticipated driven by these forthcoming technological changes? How will they impact the use of these systems?
[Johannes] Faster line scan cameras will impact this technology. This means higher throughput and improved sorting quality and these together with simpler integration and installation tools (remote control, data mining) due to the impact of the IT market.

[Joerg] New sensors will extend the range of applications, for example foreign material recognition in plastics. Especially in recycling, more material will be recoverd from waste and will be reused instead of landfilling or burning.

11. Are there market changes that are driving the adoption of machine vision-based plastic pellet sorting systems? More materials to sort? More colors to sort? More recyclable plastics? Adoption of plastic containers by more markets?
[Joerg] Well, we expect a growing market based on higher quality requirements and more colors/products/quantities. The quality of high tech products is more and more important and commands a premium in the marketplace. Even recycling products will have a higher value driven by the increasing price levels for raw materials.

[Johannes] In today's competitive and global environment, all sectors that use sorting technology demand unblem¬ished, consistent product quality. By the way, the same applies to surface vision systems. Faulty prod¬ucts, discolorations, damaged products or cont¬aminants are more or less market killers.

12. Are there market changes that will require changes in the machine vision technology? What impact do these market changes have on the technical requirements (specifications) for the machine vision technology? And how will these machine vision systems have to change to address these more demanding requirements?
[Johannes] State-of-the-art are camera and laser based systems. Due to the increasing demands more and more multisensor systems (e.g., IR, UV, X-ray) are under development. This leads to another understanding of image processing. Features from different sensors have to be combined with the goal to extract additional and better features for the classification and sorting to improve the sorting quality and purity.

[Joerg] CommoDaS has already applied various sensor principles such as true colour CCD, electromagnetic, Near Infrared (NIR) and X-Ray. In more than 430 installations and another 900 NIR systems have been installed by our sister company TiTech VisionSort.
In 2001 CommoDaS received a technology award for the first sorting equipment combining electromagnetic and true colour sensors. Multisensor systems will be the future trend. Vision systems need to be prepared to process the data of different sensors with different resolution. Such complex systems need more software support for advanced teach-in and optimizing functions.

13. As a supplier of machine vision-based plastic pellet sorter systems what are some challenges you face in marketing these machine vision systems?
[Joerg] Sensor based sorting is still not very well known in the industry and people are concerned about reliability. The challenge lies in achieving a much higher acceptance of this exciting technology.

[Johannes] The general answer is costs and competition but our opinion is to understand the specific customer’s needs and most important the expectations and their added value.

14. What advice would you give to a company investigating the purchase of a machine vision-based plastic pellet sorting system?
[Johannes] Besides all technical features (throughput, sorting quality, purities, cost, added value), it is important that the customer get tools to do his job better and makes his job easier. Therefore, simple integration is important, customizable and intuitive operation and rapid product changes with software which is easy to use and a classifier “click and learn”. My advice is to make an informed decision and try to get all information to understand the principle, operation and of course the benefit in your respective environment.

[Joerg]: In addition to the machine vision solution, the sorting equipment should consist of a well designed feeding system, feed presentation and separation solution. The customer should ask for references in the industry and visit an installation. He should run a separation test with his own material in order to verify the separation performance.