Title: Temporal Deep Learning Architecture for Forecasting Melt Properties in Extrusion-Based Manufacturing Oral: Temporal Deep Learning Architecture for Forecasting Melt Properties in Extrusion-Based Manufacturing

In polymer extrusion within the manufacturing industry, melt pressure and melt temperature are critical indicators of product quality and process safety. Real-time monitoring and control of these parameters, however, are challenging due to the complexity of manufacturing industry data, which involves diverse process variables and non-linear dynamics. While deep learning and machine learning algorithms hold promise for overcoming these challenges, current literature has yet to explore their application in forecasting polymer melt properties over extended forecast horizons. In this work, we introduce a deep learning architecture that combines Temporal Convolutional Networks (TCN) for capturing short-term dependencies with a Multi-Input Stacked LSTM for modeling long-term trends. Additionally, a repeat vector is incorporated to facilitate multi-step forecasting over 60-time horizons. Experimental results demonstrate that our proposed model outperforms traditional statistical models, including multivariate Vector Autoregressive Integrated Moving Average (VARIMA), univariate Seasonal Autoregressive Integrated Moving Average (SARIMA), and linear regression based on Stochastic Gradient Descent (SGD), highlighting its effectiveness for complex, real-time applications in polymer extrusion.