Soft Sensor Design via Switching Observers

Area of Science:

• Chemical Engineering
• Control Systems Engineering
• Process Systems Engineering

Background:

• Soft sensors are vital for real-time process monitoring, fault detection, and isolation in industrial applications.
• Observer-based techniques offer a robust framework for soft sensor design, particularly for nonlinear systems.
• Accurate estimation of unmeasured variables is essential for effective process control and safety.

Purpose of the Study:

• To design advanced soft sensors for single input single output (SISO) nonlinear processes.
• To develop a general observer-based approach for nonlinear systems with known dynamics but unknown physical parameters.
• To enhance process monitoring and fault detection capabilities through accurate estimation of non-measured variables.

Main Methods:

• A novel approach utilizing identified input-output (I/O) linear approximants around operating points.
• Development of a bank of switching linear observers based on I/O measurements and process characteristics.
• Formulation of observer-oriented target areas and the dense web principle for observer selection.
• Design of a data-driven rule-based system for seamless switching between observers.

Main Results:

• The proposed soft sensor design effectively estimates non-measured process variables using identified I/O approximants and real-time measurements.
• The observer bank, tailored to linear approximations and target operating areas, ensures robust performance.
• Simulation results for a chemostat model demonstrate the satisfactory performance and efficiency of the developed soft sensor scheme.

Conclusions:

• The developed soft sensor design provides a reliable method for monitoring and fault detection in nonlinear processes.
• The approach based on switching linear observers and I/O linear approximants is general and efficient.
• The study validates the effectiveness of the proposed soft sensor through simulations, highlighting its practical applicability.