New publication on sensor-based particle mass prediction in "Waste Management"

  Methodical depiction of the three phases of sensor-based particle mass prediction Copyright: © ANTS Sensor-based particle mass prediction

Sensor-based material flow characterization (SBMC) methods are the subject of current research and have the potential to optimize future sorting plants through new applications such as automated monitoring of product qualities or adaptive process controls. In order to convert area- or volume-based sensor data into reliable material flow characteristics in mass percent, it is necessary to obtain information about individual particle masses.

A recent publication by ANTS presents a solution for sensor-based particle mass prediction through machine learning algorithms. Based on a created dataset representing the 3D shape and masses of 3,830 particles from light packaging waste, different machine learning models were trained, and the accuracy of their predictions was compared. As a result, the machine learning models studied were found to provide up to 43% more accurate predictions (R² = 0.763) compared to state-of-the-art methods (mean grammages; R² = 0.533).

Free access to this article is available until December 15, 2015, via Science Direct.