Student's-t Mixture Regression-Based Robust Soft Sensor Development for Multimode Industrial Processes.

Area of Science:

• Industrial process monitoring
• Statistical modeling
• Soft sensor development

Background:

• Industrial processes often operate in multiple modes, complicating analysis.
• Sensor data frequently contains outliers, hindering accurate statistical analysis and soft sensor development.

Purpose of the Study:

• To propose a robust soft sensor approach for multimode industrial processes with outlier-contaminated data.
• To address the challenges posed by process variability and data imperfections in soft sensor accuracy.

Main Methods:

• Student's-t mixture regression (SMR) is proposed, utilizing Student's-t distributions instead of Gaussian distributions for modeling.
• The SMR explicitly considers the functional relationship between secondary and primary variables.
• A computationally efficient parameter-learning algorithm is developed for the SMR.

Main Results:

• The proposed SMR approach demonstrates effectiveness in developing robust soft sensors.
• The method successfully handles processes with multiple modes and outlier-contaminated datasets.
• Validation on a numerical example and a real-life industrial process confirms the approach's feasibility.

Conclusions:

• Student's-t mixture regression offers a robust solution for soft sensor development in challenging industrial environments.
• The developed SMR method improves accuracy and reliability in the presence of process modes and data outliers.