Making Recycling More Cost-Effective with AI: Insights from NIST Research

Recycling can be a significant expense for local governments, but AI might help cut those costs and potentially increase recycling rates. Researchers at NIST are working on making recycling more efficient and less expensive.

Ever wondered what happens to your plastic after you toss it in the “recycle bin”?

This question has been popping up in the news a lot recently.

The answer is quite complex. It depends on where you live and what type of plastic you’ve thrown away.

Collecting recyclables is a huge cost for local governments. They need to maintain facilities to process plastics, as well as trucks and bins to collect them. They also need to hire people to do the job. It would be much cheaper to just dump everything in landfills.

However, when local governments recycle, they can turn trash into cash if they have the right infrastructure. They can offset some costs by selling collected plastics back to manufacturers. Most manufacturers want recycled plastics to be almost as good as new, which requires careful sorting to provide consistent products.

To most people, all plastics look the same. But keen eyes know there are seven common types of plastic. You can identify them by the small recycling symbols on the bottom of almost all plastic containers. These numbers help identify the chemical makeup of those plastics. You might have noticed them when sorting your own recycling.

Here’s a breakdown of some of these materials:

Sorting these plastics is crucial. Different types of plastic with similar characteristics often can’t be mixed because they require different melting processes.

Take PVC, for example. It’s used in everything from pipes to shower curtains. Melted PVC produces a strong acid useful in many industrial applications. But, like many other acids, it’s not something you want to make unexpectedly.

Polyolefins, a group of plastics including HDPE (used in milk jugs), LDPE (used in plastic bags), and PP (used in takeout containers), provide a milder example. These plastics make up about 40% of the world’s plastic production. They are also some of the hardest to sort.

The type of plastic used in milk jugs requires high temperatures to melt and reprocess due to its crystalline structure. However, if plastic bag contaminants get mixed in, those bags degrade at these high temperatures. So, if a plastic bag gets mixed in with milk jugs, it could result in a batch of off-color, unusable milk jugs. This processing risk is one of the reasons you don’t see many milk jugs made from recycled plastic.

Additionally, if some high-temperature-stable materials from takeout containers end up on a plastic bag processing line, you could see machine clogs.

Workers at the Montgomery County Recycling Center sort materials for recycling.

In theory, you can easily sort plastic waste by using the small recycling symbols. Then, you can sell these sorted plastics to secondary recyclers, who turn them into products.

The price depends on the assumed purity of the plastic. A big bundle of orange detergent bottles might sell for a high price because they are easy to pick out. However, a pile of takeout containers might easily get mixed with various colors or additives.

At the local recycling facility in Montgomery County, Maryland, people manually sort detergent bottles, food containers, and more. However, hands and eyes can only move so fast, and it’s easy to make mistakes at that speed. So, recycling facilities focus on sorting high-value or easy-to-identify plastics to maintain consistency when selling to secondary recyclers. This means detergent bottles and beverage containers get recycled at high rates. Your plastic “cutlery” and old children’s toys might not.

To aid sorting, our work at NIST has focused on using Near Infrared (NIR) light, a technology that can quickly identify different plastics. Some top recycling facilities already use lights or cameras to “see” and sort soda bottles from PVC pipes.

But these systems can’t sort everything. My research focuses on creating a method to help sort the most challenging plastics so recyclers can turn a profit.

How We’re Making Recycling More Efficient

With this in mind, our team looked at this NIR method and decided to improve it with machine learning algorithms and other scientific techniques.

In infrared spectroscopy, you shine different wavelengths of light on some molecules. These molecules absorb part of the light’s energy at specific wavelengths and reflect or transmit the rest.

One way to think about this is with flowers and colors. For example, when many wavelengths of light from the sun shine on a red rose, the rose is very good at absorbing every wavelength/color except red. The red light reflects off the petals, which is why the rose appears red to us.

If we know the color and intensity of the light we’re shining on a flower or plastic bottle and the color/intensity we get back, we can use the differences to identify more of these flowers or bottles, like a fingerprint.

Using machine learning, we can find the NIR fingerprints of many plastic materials. We then “train” computers to identify plastics based on new NIR signals compared to other plastics’ NIR signals. This training helps the technology recognize materials in soda bottles, understand how they differ from takeout containers, and separate them accordingly.

In our first paper, we used machine learning to connect our plastic signals to certain properties (like how dense and crystalline polyethylene is). Typically, you measure density by weighing plastic in different liquids and comparing differences. It’s a very slow and tedious process.

However, we showed that you can find almost the same information using reflected light—much faster. On a recycling line, time is crucial.

You can apply this method to large and small samples. This is cool because it shows that if we set things up carefully, we can get more information from these light-based measurements.

This is still very preliminary work and doesn’t apply to all types of plastics yet. So, we can’t just shine light on any plastic and know its exact properties, but it’s an exciting start. If we can scale it up, it could save recyclers and manufacturers a lot of time and effort in quality control steps.

Since publishing this work, I’ve been delving into how to handle all the data from these measurements. You end up with very different data based on the shape of the plastic and whether the sample is a pellet, powder, or bottle.

This is because light still reflects, but it reflects in different directions depending on the plastic’s shape. Imagine the reflections on a clear pond versus a pond with many ripples. Then, you can add pigments and preservatives that might really change the signal. This doesn’t make the data wrong, but it can affect sorting. You can think of it as categorizing photos of people in black and white versus the same people in black and white, color, comics, and paintings.

To tackle this, the team has been expanding our dataset, and I’m looking at mathematical fixes to put powders, pellets, and colored plastics on the same playing field. If we can do this, identifying which plastic is which using machine learning becomes easier.

To make this research more broadly useful, I’m working to show we can sort those tricky polyolefins. Using my current method, we’ve reached 95% to 98% accuracy in sorting these plastics. We’re doing this with processes that almost any recycling facility equipped with NIR can quickly start using.

Many recycling facilities might already be using similar algorithms, but this work provides an extra level of refinement, focusing specifically on difficult-to-sort polyolefins.

If we can effectively sort these, we can reuse them with fewer processing problems, making recycling more profitable. Then, hopefully, profits can drive better recycling habits, and we can start turning our linear economy into a circular one.

Recycling as a Puzzle to Be Solved

I’m a problem solver, jumping from one puzzle to the next.

Besides polymer research, I’ve worked on drug delivery systems for ovarian cancer, and now I’m using artificial intelligence (AI) and machine learning.

I love doing good while solving complex problems. Sustainability and bio-friendly materials have been a beautiful theme throughout my research career.

You might not initially see the connection between biomedical research and plastics. But drug delivery systems can help create really cool materials with applications beyond medicine. Plastic work can also enhance our understanding of DNA, proteins, and collagen in our bodies.

Now, with the explosion of AI, we have new tools to do materials research faster and more efficiently. It’s an exciting time in the field of sustainable materials!

The Future of Sorting Research

I’m currently finishing a two-year contract at NIST and looking for the next puzzle to solve.

However, I plan to stay connected with NIST as an affiliate to help other researchers use my techniques.

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