How AI can help reduce landfill waste

As a provider of electricity, water cycle management and heating services, and as Italy’s largest waste management and recycling company, Hera is on the front lines of today’s battle to reduce waste and minimize environmental damage. The company has a reputation as an innovator, and it is bringing a forward-looking spirit to environmental stewardship.

Andrea Bonetti, Hera’s Manager of IT Architecture, explains, “Hera has adopted, with absolute conviction, the circular economy.” Where traditional recycling practices may be one arc in the cycle of re-use, Hera offers integrated solutions that help complete the circle. With plastics, for example, it not only recovers waste but also incorporates it into production of high-quality new products that are themselves recyclable. “Today, in our territories, most of the waste is recovered,” says Bonetti. “Only a small portion ends up burnt, but this is burnt in waste-to-energy plants, producing new energy.”

The recovery process, of course, depends on quickly finding and separating reusable material from vast quantities of refuse. It was this process where Bonetti and his colleague, Alessandro Collina, Hera’s Head of IT Innovation, decided to explore how AI-powered automation could improve efficiency and help channel more material to new use.

The challenge is twofold. Evaluating the potential of AI for waste sorting is one part. The other part is having the flexibility to incorporate this kind of innovation and scale it from laboratory to enterprise dimensions.

Currently, Hera personnel analyze waste manually. As trucks unload at the entrance to the plants and the trash is pushed toward conveyors, spotters watch for recoverable materials—including plastics, glass, aluminum and organic material—and help direct downstream sorting. It’s an onerous job on its own; but consider it at scale: 1,400 spotters across 89 plants. 6.3 million tons of waste treated every year. In other words, there is potential for much greater efficiency.

The vision is to capture video of incoming trash and have AI recognize characteristics of items and materials that would qualify them for recovery and reuse. “This could have a decisive impact on the costs of recovery and disposal activities, which is the focus of the circular economy,” explains Bonetti.

To realize this vision, Bonetti explains, “We needed a partner who could really go beyond a proof of concept and facilitate the project with both working methodologies and effective tools. We thought we could find all this in the IBM Garage.”

For the user-centered and collaborative IBM Garage team, the first step toward a solution was a design thinking workshop that involved gaining firsthand experience of the working environment. “The guys at the Garage had to ‘get their hands dirty’—a particularly fitting metaphor in this case,” says Bonetti. “We have learned from experience to immerse machine learning specialists in reality, which is always much more complex than the laboratory. So the Garage team came to see a plant. Waste is, by definition, deformed and crowded, and the lighting conditions are variable. It is not like recognizing kittens in Facebook photos!”

In fact, the Hera and IBM Garage teams quickly recognized that the plants were not the right place to capture video. There was too much material going by in too little time. Instead, they identified a better vantage point upstream.

By mounting cameras on trash trucks, they could video the smaller amounts of material falling out of bins. “It’s still an extremely rapid passage of images,” says Bonetti. “But the study of these images has allowed us to identify significant patterns for the qualitative evaluation of the waste during the collection process, not inside the plant, which could improve the time and cost of the transformation process.”

In addition, the Hera team hopes to correlate waste-quality data with collection locations, helping the company develop targeted information campaigns to help people better differentiate between waste items.

Following the agile IBM Garage Methodology, in eight weeks Hera and the IBM Garage team co-created and released a minimum viable product (MVP) that incorporates IBM Watson® Studio and IBM Watson Machine Learning technology to generate a specific tool for the use case, including a machine learning model to recognize the key waste patterns. According to Collina, “The IBM tools allowed us to take thoughts that previously were only written on paper and make them reality in a much faster and more agile process.”

Collina continues: “Now, the most urgent challenge is to understand how this can be industrialized. How can we do a prototype, for example, on a single truck for perhaps a year, with all the variations in lighting and weather, and continue getting the right insight throughout the year.”

The work Hera is doing with IBM Garage and the IBM Cloud Paks and WebSphere solutions is not a finite project; it is part of a cycle.

As Bonetti, Collina, and their colleagues work with the IBM Garage team to learn how to train AI to find recoverable waste, and how to scale such an innovation to the everyday environment, they will not only cultivate new and valuable expertise; they will also identify other ways to put AI to work in the utilities industry.

The modernization and flexibility Hera has brought to Beam IoT instills greater reliability and resilience in a critical system while opening it up to ongoing adaptation and expansion.

And most important, because these efforts could drastically improve the cost efficiency of reclaiming waste for other uses, they have the potential to influence an industry, generating greater momentum in the circular economy and allowing the world to reclaim some of its green.

Get started with IBM Consulting

Hera (link resides outside of ibm.com) is a leading multi-utility in Italy, providing energy distribution, energy sales, water cycle management, heating, and waste and recycling services. It operates across the Emilia-Romagna, Veneto and Friuli regions, employing more than 8,000 people and earning more than €6 billion in yearly revenue.