A novel automated soft sensor design tool for industrial applications based on machine learning

Abstract

In modern industrial processes, real-time monitoring and control of key quality variables are crucial but challenging due to measurement limitations and process complexities. Traditional methods for developing soft sensor models are not only time-consuming and labor-intensive but also require substantial expertise in machine learning, and often lack user-friendly interfaces, thereby limiting their accessibility to engineers in the field. To address these issues, this paper introduces an easy-to-use, open and efficient automated soft sensor design tool called Soft Sensor Manager. The Soft Sensor Manager incorporates advanced supervised, semi-supervised, and causal machine learning algorithms to enable effective model development and deployment. It also provides functionalities such as data preprocessing, feature engineering, algorithm selection, hyperparameter optimization, model evaluation and online deployment within a user-friendly interface. The software’s effectiveness was demonstrated through its application in predicting light catalytic cracked oil yield using real industrial data. By automating the soft sensor design process, the Soft Sensor Manager enhances modeling efficiency and model quality, ultimately contributing to improved process monitoring and optimization in industrial settings.