Adaptive soft sensor considering process state in film manufacturing process and identification of critical process variables

Abstract

In the film manufacturing process, process variables, such as temperature and pressure, are measured and controlled to manage the film properties, such as thickness and optical characteristics. Each film property is regulated by the product specifications, and having all film properties adhere to the specifications is important. In this study, soft sensors were utilized to predict film properties and the importance of the variables that affect the film property variations for each soft sensor was calculated. Different process states were considered in the soft sensor modeling, a different soft sensor was constructed for each state, and soft sensor models with film property prediction accuracies higher than those of the conventional ones were achieved. The importance of the process variables was calculated by adding random variables to identify the truly important variables. The appropriate process variables were confirmed to be detected from the perspective of on-site process engineers. This study contributes to the application of soft sensors in film manufacturing processes.

Keywords

• Soft sensor
• Film properties
• Process variables
• Machine learning
• Manufacturing process
• Regression analysis