Abstract
The study leverages Machine Learning (ML) techniques to optimize the control parameters of a multivariable Proportional-Integral (PI) controller for a distillation column. The methodology involves generating data through MATLAB simulation, utilizing TreeNet regression for predictive modeling, and performing multi-response optimization to minimize Integral Time Absolute Error (ITAE) for both top and bottom tray temperatures. The TreeNet models achieved R² values of 95.10% for ITAE_e1 and 93.85% for ITAE_e2, indicating high predictive accuracy. The resulting optimal controller settings were: Kp11 = 1.6859, Kp12 = − 2.0610, Kp21 = 3.1846, Kp22 = − 1.2176, Ki11 = 1.0628, Ki12 = − 1.2989, Ki21 = 2.4540, and Ki22 = − 0.7676. Compared to traditional control techniques such as Davison’s PI and IMC, the ML-optimized controller significantly reduced ITAE, improved stability, and enabled faster system responses. This study contributes a data-driven and interpretable framework for PI tuning in MIMO systems, with practical implications for process industries aligned with Industry 4.0.
| Original language | English |
|---|---|
| Pages (from-to) | 1749-1765 |
| Number of pages | 17 |
| Journal | International Journal of Advanced Manufacturing Technology |
| Volume | 146 |
| Issue number | 3 |
| DOIs | |
| State | Published - Sep 2026 |
Keywords
- Distillation column
- Industry 4.0
- Machine learning
- Multi-response optimization
- Proportional-integral control
Fingerprint
Dive into the research topics of 'Machine learning-enhanced optimization of PI controllers for multivariable distillation columns'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver