Methods of digitalization and artificial intelligence (AI) can make a major contribution to improving waste recycling. Within the AI-Waste research project, the recycling share will be increased by at least 10 % through innovative approaches. Image recognition and machine data analysis will be combined to optimize waste processing.

Mountains of waste and plastic are growing continuously worldwide. Around 4.4 million tons of municipal waste from private households and similar establishments are generated in Austria every year. This is mainly mixed waste, consisting mostly of plastics and composites as well as organic fractions such as paper and cardboard. For waste processing, the diverse composition is challenging since it varies greatly seasonally and regionally. Existing plants do not have a widely used or suitable technology to automatically detect the quality of intermediate steps at a plant. As a result, it can happen, for example, that the plastic bottle fractions is well separated, while the remaining waste components such as cardboard packaging are insufficiently separated.

Technological milestone

The AI-Waste project is now combining image data with plant data for the first time in order to describe the type and composition of waste in the ongoing process. The project, which is led by Know-Center, is being implemented in collaboration with the Institute for Machine Vision and Display at Graz University of Technology and JOANNEUM Research Forschungsgesellschaft.

“Digitalization offers untapped potential, especially in our area of activity. Constant innovation is the only effective means of achieving long-term success as a company,” emphasizes Christian Oberwinkler, CTO of Komptech GmbH, which supports the project as a technology partner in the area of mixed waste processing.

Detect and distinguish waste

To ensure that waste is described correctly for subsequent shredding, the material stream must be divided into individual objects. At the same time, information about class affiliation and geometry of objects is necessary in order to classify them unambiguously. Objects of the same class, such as wood pellets and branches, have different geometries. Objects with similar geometry, such as PET bottles and glass bottles, are in turn assigned to different classes. If, ideally, each object is recognized, the material distribution on a recorded surface can be determined and the recycling machine can be adjusted accordingly.

3D sensor technology, such as stereo cameras, time-of-flight (TOF) cameras, is used to capture the spatial properties of objects. 2D sensors describe the color information with very high resolutions. The image analysis software applies deep learning algorithms, which learn to recognize and distinguish waste based on training data. In terms of better understanding the interrelationships of plant operation, the research team examines influencing factors and correlations within the measurement data. Based on this, models are derived to optimize the plant.

Profit for companies and environment

The result of AI-Waste will be a recommendation for action on how AI can be used in process optimization for waste and recycling management. As a result, waste management companies will benefit from increased efficiency, increased recycling rates and reduced energy consumption, which will subsequently have a positive impact on the environment. Overall, the project contributes to optimizing image recognition in terms of measurement accuracy and position.

The results will also serve as preliminary work for other industrial sectors, such as pharmaceutical and steel industries, where image data also must be analyzed together with time series data.

 

The project KI-Waste is funded by the Zukunftsfonds Steiermark and the Klimafonds Graz.