Developing a soft sensor Random Forest model for the inline product characterization of Polylactide (PLA) in a twin screw melt extrusion process

摘要

The melt processing of Polylactide faces challenges due to its poor thermal stability which is influenced by processing temperatures and shearing. The characterization of processed products takes place offline in laboratory environments. Typical scrap rates of a medical grade product can be up to 25-30%. This work discusses the development of soft sensor random forest models for a twin screw melt extrusion process. The resulting models can predict product end characteristics from inline data. These include mechanical properties and percentage mass change of a product during its degradation cycle. These models will act as novel inline indicators as to whether products will be in or out of specification. This will reduce manufacturing costs and minimize waste as well as accurately predicting future performance and behavior of products.