Closing material and resource cycles is an integral part of the circular economy. Many products (especially electronic devices) are becoming increasingly complex regarding their structure and raw materials. To recover raw materials, the products must be broken down into ever-finier grain sizes and subsequently sorted. In general, dry sorting processes achieve good sorting results in the fine particle size range with lower throughputs. However, when increasing the throughput, the quality of the sorting results decreases considerably. In order to operate profitably, sorting machines must work as close as possible to the tipping point between good quality and maximum throughput while also minimising maintenance downtimes.
The SoRec project focuses on digitalising sorting processes for fine-grained, metal-containing waste streams in the recycling industry. By installing state-of-the-art industrial line cameras and sensors, we are digitizing the conventional sorting method on a moving belt. With the help of advanced AI models and algorithms in deep learning and machine learning, our system can accurately detect materials on the conveyor belt, classify them based on size, shape, and color, and even find their precise edges. With the capability to identify multiple layers of materials, the AI model provides valuable density and volume estimation. To ensure real-time efficiency and control, we have integrated the AI model with powerful computer vision techniques, which handle crucial image processing tasks. This seamless collaboration between AI and computer vision allows us to estimate the belt’s speed and detect any anomalies, ensuring precise sorting and preventing belt misalignments. Our materials, measuring just 1 mm in size, demand meticulous attention to detail, necessitating high-level zoom capabilities for precise annotations. With this innovative AI-driven system, we are taking a significant step towards automating and optimizing the sorting process, enhancing productivity, and elevating the industry to new heights of accuracy and efficiency.