Black box modeling of PIDs implemented in PLCs without structural information: a support vector regression approach

Area of Science:

• Automation and Control Systems
• Machine Learning Applications
• Industrial Process Optimization

Background:

• Proportional-Integral-Derivative (PID) controllers are ubiquitous in industrial automation.
• Tuning PID controllers in programmable logic controllers (PLCs) is challenging due to proprietary manufacturer modifications and lack of structural information.
• Existing methods often require detailed knowledge of the PID algorithm's internal structure, which is frequently unavailable.

Purpose of the Study:

• To present a novel method for parameter identification of PID algorithms within PLCs.
• To utilize Support Vector Regression (SVR) for modeling PID controllers as black boxes.
• To demonstrate the effectiveness of SVR in accurately mapping PID functions, including manufacturer-specific modifications.

Main Methods:

• Implementation of a Support Vector Regression (SVR) model for PID parameter identification.
• Focus on feature selection to enhance model training and testing capabilities.
• Testing the SVR-based method on real-world PLCs from Siemens and General Electric.

Main Results:

• The SVR model accurately represents the behavior of PID algorithms and manufacturer modifications.
• High accuracy achieved in mapping the functional behavior of the PID controllers.
• Simulation results derived from the SVR model can be directly applied for PID tuning.

Conclusions:

• The proposed SVR-based approach offers a universal solution for tuning PI/PID controllers in PLCs, regardless of proprietary modifications.
• This method provides a valuable alternative for engineers lacking detailed information on PLC PID structures.
• Enables effective PID tuning even when default settings for known structures cannot be used.