Virtual Sensor for Sustainable Large-Scale Process Monitoring

Feed quality diversity, high production capacity, and often harsh process conditions lead to challenges in process monitoring and control. This study proposes a framework that advances soft (virtual) sensor applications in complex industrial settings, highlighting the potential of machine learning and process knowledge for process monitoring and digitalization by utilizing a plantwide process control system. The application is demonstrated through a soft sensor to predict Brix values as a quality indicator variable (QIV) in industrial batch crystallization processes. The effectiveness of the soft sensor is shown through a rigorous evaluation using statistical metrics. Additionally, the study explores temporal dependencies through autocorrelation function (ACF) analysis and long short-term memory (LSTM) modeling. The developed soft sensor is applied to a sugar manufacturing plant. It has been shown that it will decrease the mean residence time of sugar crystallization by 2.12%, which consequently leads to a reduction in energy consumption and CO2 footprint, increasing production capacity, and finally, sustainable process development.