Digitalization and Industry 4.0 are in their early stages of implementation within the waste management industry. In the future, real-time monitoring, data analysis, management and process control for waste treatment plant management will be the top priorities for development. Therefore continuous advancements in technology and procedures are necessary for effective and efficient operations in this domain (Sarc and Pomberger, 2022). Mixed waste treatment presents challenges, including the variable quality of waste and fluctuations caused by material and machine-related factors.

 

The primary objective of this study is to characterize solid waste particles within a bulk on a conveyor belt, focusing on bulk processing. Real-time characterization of waste on conveyor belts is a fundamental necessity for plant optimization in real-time. The aim of this study is to achieve autonomous and dynamic adjustments to cope with waste fluctuations. While possible, necessary sensors for comprehensive characterization come with considerable costs and demanding in terms of work safety. Therefore, the potential of classification through artificial intelligence and cameras becomes significantly desirable. However, these methodologies demand substantial volumes of high-quality labeled training data, which proves intricate to obtain for waste materials in the context of bulk processing. Thus, the creation of such training data through digital means, such as 3D renderings, is being contemplated.

 

This work stems from the larger KalKIDEM project - Calibration of Discrete Element Simulation Models using Particle Sensor Data and Artificial Intelligence. In contrast to the simulation target of the project, this initiative pursues the approach of achieving a less computationally intensive replication of particles. The concept revolves around simulating particles using a physics engine, capturing realism without the complexity of extensive calculations typical in DEM simulation programs.

In this regard, the initial particle simulation utilized a well-established game engine, Unreal Engine 5.2.1, recognized for its proficiency as a physics engine (Kose et al., 2017).

The study has encountered preliminary challenges, primarily focused on the generation of coherent 3D models. Initial efforts are directed toward devising an optimal setup, wherein particles are manipulated within 3D modeling software and achieving a realistically rendered bulk, thereby discerning mass composition. The overarching aim is to make use of machine learning techniques, furthering the enhancement of plant performance.