Minimizing Temperature Deviations in Rubber Mixing Process using Artificial Neural Networks

Abstract

Rubber mixing is a complex manufacturing process that poses challenges for process control due to the high number of control variables including mixing parameter settings, rheological behaviour, compound viscosity and batch-dependent material variations. Already small deviations from the control variables can influence the compound properties, leading to increased scrap rates. To address these challenges, this paper introduces an Artificial Intelligence (AI)-based approach to enhance process control in rubber mixing by predicting mixing temperatures from input variables. The proposed method utilizes Feedforward Neural Networks (FFN) to enable early identification of batch-specific temperature deviations, enabling systematic improvements with each new application. The FFN was trained on a diverse dataset encompassing various rubber recipes and batches. Post-training, the FFN demonstrated remarkable accuracy, achieving a Mean Absolute Percentage Error (MAPE) of 1.00% on the training dataset and 1.44% on the validation dataset, thereby showcasing its efficacy in predicting temperature fluctuations within the mixing process. Consequently, the FFN can determine the relevant input variables necessary to achieve specific mixing temperatures, providing a foundation for an automated control system in rubber mixing process. This paper outlines the system architecture of the FFN tailored for rubber mixing and provides a comprehensive overview of the experimental results.