In rubber extrusion, precise temperature control is critical due to the process’s sensitivity to fluctuating parameters like compound behavior and batch-specific material variations. Rapid adjustments to temperature deviations are essential to ensure stable throughput and extrudate surface integrity. Based on our previous research, which initiated the development of a feedforward neural network (FNN) without real-world empirical application, we now present a real-time control system using artificial neural networks (ANNs) for dynamic temperature regulation. The underlying FNN was trained on a dataset comprising different ethylene propylene diene monomer (EPDM) rubber compounds, totaling 14,923 measurement points for each temperature value. After training, the FNN achieves remarkable precision, evidenced by a mean absolute percentage error (MAPE) of 0.68% and a mean squared error (MSE) of 0.63°C2 in predicting temperature variations. Its integration into the control system enables real-time responsiveness, allowing for adjustments to temperature deviations within an average timeframe of 68 ms. A key advantage over proportional-integral-derivative (PID) controllers is the ability to continuously learn and adjust to complex, non-linear, and batch-specific process dynamics. This adaptability results in enhanced process stability, as evidenced by inline manufacturing validation. Our paper presents the first ANN-based rubber extrusion control, demonstrating how machine learning techniques can be effectively leveraged for real-time, adaptive temperature control. Beyond rubber extrusion, this strategy has potential applications in various polymer processing and other industries requiring precise temperature control. Future trends may involve the integration of online learning techniques and the expansion of interconnected manufacturing processes.
