好色先生TV

Despite efforts to toss empty soda cans and take-out containers into recycling bins, most of the plastic people use every day聽. That鈥檚 largely because plastic needs to be sorted before it can be recycled 鈥� and the process is costly for recycling facilities.

A team of students from 好色先生TV鈥檚聽 investigated whether recycling facilities could effectively sort plastics by the sounds they make when dropped, as density differs by plastic type. They worked with the聽, a Pittsburgh-based company that develops technologies to lower the cost of recycling.

By the end of the fall semester, the students had created a machine-learning model that can classify plastic types by sound with over 90% precision. Their work addresses a need in the recycling services sector, as current sorting methods are not only expensive but also can generate waste and misidentify plastic types.

The team took on this challenge as part of their聽, which allows students to develop solutions for real-world clients with skills gained from Heinz College. The Reclamation Factory team included Akshay Kulkarni, Dhivya Sreedhar, Bryan Tan, Yvonne Tang and Abhi Vakil, all students in the聽 辫谤辞驳谤补尘.听

鈥淚 really think this was one of those projects that truly screams 鈥楳ISM鈥� because it has everything in it, from the science perspective, to building the entire machine-learning model, and looking at the business side as well,鈥� Kulkarni said.

Discovering the sounds of plastic

Recycling the tons of plastic thrown away each year often does not make financial sense for recycling facilities, according to NPR. There's a limited market for recycled plastic because new plastic is cheap to produce, which means the complex and pricey process of sorting old plastic for reuse often generates little return on investment.

Reducing the cost of sorting, however, could help recycling facilities repurpose more old plastic instead of sending it to a landfill. That鈥檚 where the Reclamation Factory comes in.

In summer 2025, Sreedhar interned with the Reclamation Factory. Tasked with finding a better way to sort plastics, she and the other interns discovered that different types of plastic emitted distinct sounds when dropped on a concrete floor. So, they continued dropping plastics, collecting audio samples from about 1,000 items across six plastic types.聽

鈥淚t was really just a very random experience,鈥� Sreedhar said. 鈥淲e were thinking, 鈥楥an we leverage this and see if the audio shows proper distinguishable features?鈥欌��

With the audio data in hand, the team鈥檚 next step was聽, which involves turning data into inputs for a machine-learning system to base predictions on. The students chose three variables that, altogether, capture the distinct sounds of the six plastic types. Then they used the inputs to train and evaluate the performance of several machine-learning models.聽

In the end, the best-performing model could correctly identify two types of plastic often found in consumer products 鈥� polypropylene and polyethylene terephthalate 鈥� with 100% precision, and it could identify a third, polystyrene, with 92% precision. The three other types of plastic were not the main focus of the project but were included to further validate the model.

The team also created a user interface for the model, allowing the Reclamation Factory to run demonstrations for its clients.

鈥淲hat I loved about this particular experience was that it had a bit of everything,鈥� Kulkarni said. 鈥淲e had to think not only about building the machine-learning model 鈥� we had to also think about the structure of these plastics. How are their physical structures? What acoustic signatures do they produce?鈥�

The Reclamation Factory was a valuable partner, students said. They met weekly with two members of the company, receiving data and literature on sorting items based on density and audio frequencies. Tan said he felt the team was working with the client to solve a problem, rather than pursuing busywork.聽

Sreedhar added that the project 鈥溾�嬧�媤asn't just about meeting the business goals and getting the metrics 鈥� Reclamation Factory made sure we all had a proper learning experience.鈥澛�

Georgia Crowther, CEO and principal robotics engineer at the Reclamation Factory who earned a master's degree in robotics systems development from CMU in 2019, said the capstone team turned the concept of acoustic sensing into a viable product.

鈥淭hey took on some really fundamental research to get these incredible results in the span of just a few months,鈥� Crowther said. 鈥淎s a startup with a small team and limited resources, I don鈥檛 think we could have gotten our technology to where it is today so quickly without their hard work.鈥�

The 鈥榤agic鈥� of collaboration

Most of the team did not have any experience with audio classification, which made the work initially feel intimidating, Tang said.

鈥淲e were venturing into uncharted territory where a lot of the terms, like 'spectrogram,' were a foreign language for us. But I think magic happened in our collaboration,鈥� Tang said. 鈥淲e had to use our collective intelligence to figure out solutions and find answers. We spent a lot of hours together, and that proved to me that a committed team can bridge any knowledge gap.鈥�

The project provided an opportunity for the students to exercise project management skills, Vakil said, as they were responsible for communicating with the client, dividing up tasks among one another and creating a safe space for sharing ideas. For Kulkarni, the project demonstrated the importance of prioritizing the user experience when creating a solution.聽

The team鈥檚 collaborative efforts helped them advance the project while handling other priorities, such as coursework, exams and job applications, Tan said. At the project鈥檚 conclusion, he had gained valuable skills in machine learning, as his background is primarily in user experience research and design engineering.聽

鈥淚 learned a lot from the other team members who were more technically proficient in the model training side of things. That was an incredibly fun experience, being able to pick things up on my own while learning from others,鈥� he said.聽

Putting skills into practice

Sean Beggs, director of the MISM program at Heinz College, said capstone projects are a valuable and rewarding experiential learning tool. They allow students to demonstrate critical thinking and problem-solving skills and learn to embrace the messiness of developing unconventional real-world solutions, he said.

鈥淥ur students are trained by the very best and brightest faculty,鈥� Beggs said. 鈥淚n addition, the students have an intelligent action mindset that's developed by leaning into the applied knowledge gained in the classroom. I am grateful to the Reclamation Factory for bringing this challenge to Heinz College!鈥澛�

The interdisciplinary nature of the Heinz College curriculum prepared Tan to take a holistic view of the Reclamation Factory project, focusing not only on the final product鈥檚 technical component but also its user experience and delivery.聽

The students on the team valued the college鈥檚 courses on negotiation, managing disruptive technologies, operationalizing artificial intelligence and professional speaking, among others. Sreedhar said the MISM curriculum鈥檚 blend of machine learning, data analysis and business helped her thoroughly explore solutions to the capstone problem.聽

Applying skills gained in their graduate studies 鈥渇or the greater good of the environment,鈥� Vakil said, 鈥渕ade this project so interesting.鈥�