From trash to cash: How AI and machine learning can help municipalities reduce recycling costs

Machine Learning


Brad Sutliff wears safety glasses as he stands in front of a computer monitor in his lab.

Recycling costs local governments a lot of money, but AI could reduce the cost of that process and lead to increased recycling. NIST research aims to make recycling more efficient and cheaper.

credit:

M. King/NIST

What happens to recycling after you throw your plastic in the “trash”?

This question seems to be in the news a lot lately (see here, here, or here).

The answer is actually complicated. It depends on where you live and the type of plastic it is.

Recycling collection costs local governments a large amount of money. Facilities to handle plastics and trucks and bins to collect them must be maintained. Governments also need to hire people to do their jobs. It could be much cheaper to throw everything in a landfill.

But when municipalities recycle, they can turn trash into cash if they have the right infrastructure. Part of the cost can be offset by selling the recovered plastic to manufacturers. Most manufacturers want recycled plastic to be nearly as good as new plastic, but careful separation by recyclers is required to provide a consistent product.

To many people, all plastics look the same. However, astute observers know that there are seven types of common plastics. You can identify them by the small recycling mark on the bottom of almost all plastic containers. These numbers help determine the chemistry behind the plastic. Some of you may have noticed this when sorting recycled items.

Some of these materials are described below.

material Common uses recycling labels
polyethylene terephthalate soda bottle, recyclable water bottle 1 – Pete
high density polyethylene milk bottle, detergent bottle 2-HDPE
PVC pipes, shower curtains 3 – PVC
low density polyethylene shopping bags, sandwich bags 4-LDPE
polypropylene Takeout containers, yogurt cups 5-PP
polystyrene disposable coffee cup 6-PS
other Safety glasses, DVDs, and plenty of reusable water bottles 7 – Others

It is very important to separate these plastics. Different plastics with similar properties often cannot be mixed because they require different melting steps.

Let's take PVC as an example. Used in everything from plumbing to window blinds, PVC produces strong acids when melted and is used in many industrial applications. But like many other acids, it's not something you want to make when you least expect it.

Polyolefins, a group of plastics that includes HDPE (used in milk bottles), LDPE (used in plastic bags), and PP (used in takeout containers), are a very mild example. This group of plastics accounts for approximately 40% of the world's plastic production. These are also the most difficult to classify.

The type of plastic used in milk bottles requires high temperatures to melt and reprocess due to its crystalline structure. However, if the plastic bag is contaminated with contaminants, the high temperature will cause the bag to deteriorate. So when plastic bags go into the recycling mix with milk bottles, you can end up with ugly yellow milk bottles that no one wants to drink from. This processing risk is one of the many reasons why you rarely see milk bottles made from recycled plastic.

Additionally, if hot, stable material from takeout containers flows into the plastic bag production line, it can cause machine jams.

Workers at a recycling facility separate recycled plastic by sorting it into individually labeled yellow bins.

Workers at the Montgomery County Recycling Center separate materials to be recycled.

credit:

B. Sutliff/NIST

In theory, plastic waste could be easily sorted using small recycling symbols. The separated plastics can then be sold to secondary recyclers, who can turn the separated waste into products.

Prices vary depending on the assumed purity of the plastic. A bale containing a large orange bottle of laundry detergent is likely to sell for a high price because it is easy to choose. However, many takeout containers can easily contain a mix of plastics with different colors and additives.

At a local recycling facility in Montgomery County, Maryland, people are manually sorting items like laundry detergent bottles and food containers. However, human hands and eyes move very quickly, and that speed makes it easy to make mistakes. As a result, recycling facilities focus on separating high-value and easily identified plastics to ensure consistency in what they sell to secondary recyclers. This means detergent bottles and beverage containers are recycled at a high rate. Plastic “silverware” and old children's toys are probably not.

To facilitate sorting, NIST is focusing on using near-visible infrared light (NIR), a technology that can see plastic and quickly tell what it is. Some state-of-the-art recycling facilities already use this approach with lights and cameras that “see” and separate soda bottles from PVC piping.

However, these systems cannot separate everything. The focus of my research is to create methods to help separate the most difficult plastics in a way that is beneficial to recyclers.

How to make recycling more efficient

With that in mind, our team decided to consider this NIR approach and improve it with the help of machine learning algorithms and other scientific methods.

Infrared spectroscopy involves shining light at several different wavelengths onto several molecules. These molecules absorb some of the energy from the light and reflect or transmit the rest based on the wavelength.

One way to think about it is in terms of flowers and colors. For example, when the many wavelengths of light in sunlight hit a red rose, the rose has an excellent ability to absorb all wavelengths/colors except red. Roses appear red to our eyes because red light reflects off the petals.

If you know what color or light intensity you shine on a flower or plastic bottle and what color or intensity comes back, you can use those differences like fingerprints to create more flowers or bottles. can be identified.

Brad Sutliff stands in his lab wearing safety glasses, showing a small plastic ring in his gloved palm.

NIST researcher Brad Sutliff works with plastic samples in the NIST lab.

credit:

M. King/NIST

Machine learning can be used to find the NIR fingerprint of many plastic materials. The computer is then “trained” to identify plastics based on how similar their new NIR signals are to the NIR signals of other plastics. This training helps the technology identify the material inside the soda bottle, recognize that it is different from the make-up of the takeout container, and separate it accordingly.

In our first paper, we used machine learning to link plastic signals to specific properties of polyethylene, such as its density and crystallinity. Density is usually measured by weighing the plastic in different liquids and comparing the differences. This is a very time consuming and tedious process.

However, we show that nearly the same information can be found much more quickly using reflected light. And time is of the essence on the recycling line.

This approach is applicable to both large and small samples. This is great because it demonstrates that with careful setup, more information can be extracted from these light-based measurements.

This is still very preliminary work and may not be applicable to all types of plastics. So while we can't shine a light on every plastic and know its exact properties, it's an exciting start. Scaling up could potentially save both recyclers and manufacturers significant time and effort on quality control steps.

Since that study was published, I've been digging into the best ways to process all the data obtained from these measurements. In the end, depending on whether the sample is a pellet, powder, or bottle, you'll get completely different appearance data based on the shape of the plastic.

This is because light is still reflected, but in different directions depending on the shape of the plastic. Consider reflections in a clear pond and reflections in a pond with many ripples. Then you can add colors and preservatives that can actually change the signal. It's not that the data is wrong, but it may affect the sorting. Think of this as the difference between classifying black and white photographs of people and classifying black and white photographs, color photographs, portraits, and paintings of the same person.

To combat this, the team is working to expand the dataset and I'm looking at mathematical corrections to bring powders, pellets, and colored plastics on the same playing field. If we could accomplish that, it would be much easier to use machine learning to identify which plastics are which.

To make this research more broadly useful, I'm working to show that those troublesome polyolefins can be classified. My current method achieves 95% to 98% accuracy in separating these plastics. We're doing this in a process that almost any recycling facility with NIR could start using very quickly if they wanted to.

While many recycling facilities likely already use similar algorithms, this effort provides an additional level of improvement by focusing on difficult-to-sort polyolefins.

If they can be separated effectively, they can be reused with fewer processing problems, making recycling much more profitable. Then, hopefully, the profits will drive better recycling habits and begin to transform the linear economy into a circular one.

Recycling as a puzzle to be solved

I'm a problem solver and jump from puzzle to puzzle.

In addition to researching polymers, he is also working on drug delivery systems for ovarian cancer and is currently working on artificial intelligence (AI) and machine learning.

I love what I can do while solving complex puzzles. Sustainability and bio-friendly materials have been important themes throughout my research career.

At first, you may not realize the connection between biomedical research and plastics. However, drug delivery systems can also help produce exceptional materials for applications outside the medical field. Research on plastics deepens our understanding of our bodies in areas such as DNA, proteins, and collagen.

And now, with the explosion of AI, new tools are emerging to make materials research faster and more effective. It's a very exciting time to be in this sustainable materials space.

The future of classification research

I'm currently finishing up a two-year contract at NIST and trying to solve the next mystery.

However, I plan to remain involved with NIST as a researcher so that other researchers can take advantage of my skills.

I want to help the larger recycling community use data analytics to improve recycling and help clean up our planet.

World Metrology Day, dedicated to sustainability, is approaching on May 20th. Using AI to help recycle more efficiently is one of NIST's many contributions to creating a healthier planet. learn more:



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *