Managing Rubber Waste Flows in Closed-Loop Production Systems: Forecasting the Mass of Vulcanized Waste Based on Technological Parameters

Rubber waste recycling is a crucial element of the circular economy. However, high uncertainty of secondary resource flows disrupts recycling stability. This study develops an organizational and methodological framework for closed-loop rubber waste management, linking organizational design with operational forecasting. A comparative analysis of organizational models justifies the cluster (centralized) model as the optimal environment for reliable data collection. Using empirical data from a single manufacturing enterprise producing rubber sealing rings by injection molding (610 shift-level observations), two forecasting models were developed and compared: a linear regression model and a multilayer perceptron (MLP) neural network. The MLP identified four significant predictors: production volume, vulcanization time, sulfur content deviation, and mold wear. Scenario analysis revealed a non-linear relationship between production and waste generation: higher production increased waste (up to 2.20 kg/shift), while extreme secondary-factor deviations (mold wear 54%; sulfur deviation 1.08%) caused waste to peak at 1.65 kg/shift even at the lowest production volume (1300 units/shift). The approach provides a transparent, interpretable mathematical expression for process engineers, shifting waste management from reactive to proactive control. Given the single-enterprise, single-product dataset, independent validation is required before wider application.