Uso de sistema de visão computacional industrial em plantas de processamento de RSU

Abstract

The proper management of Municipal Solid Waste (MSW) remains a significant logistical and environmental challenge, with efficient waste sorting a critical requirement for ensuring the safety and operational reliability of waste-processing facilities. The presence of rigid and flammable objects, such as liquefied petroleum gas (LPG) cylinders, within the waste stream transported by industrial conveyor belts poses a severe operational hazard, potentially damaging shredding equipment, endangering personnel, and causing production downtime. This work proposes a technological intervention that integrates computer vision with industrial automation for the real-time detection of such hazardous objects. The research is conducted as an exploratory, descriptive case study at an MSW processing facility, complemented by a literature review. The field study identified the spatial and temporal constraints of the processing equipment, while the literature review consolidated state-of-the-art performance metrics regarding the mean average precision (mAP) and inference latency of convolutional neural networks deployed on edge computing platforms. Based on these findings, an event-driven safety interlocking architecture between the vision system and a Programmable Logic Controller (PLC) is proposed, and its logic is verified through virtual commissioning in a Software-in-the-Loop arrangement. The proposed architecture demonstrates the mathematical and kinematic feasibility of stopping the conveyor belt before a detected object leaves the safe manual-removal zone, thereby improving operational safety while minimizing process interruptions.