This dissertation proposal introduces an evolving neuro-fuzzy model-based design of experiments for the identification of nonlinear dynamical systems. Traditional model-based process monitoring and soft sensors require frequent recalibration due to system drift, sudden shifts, and process complexity. The proposed research will investigate how evolving systems, capable of online learning from data streams, can adapt both their structure and parameters to improve robustness and reduce manual intervention. By integrating evolving neuro-fuzzy inference with adaptive design of experiments, the study aims to develop methods for automated selection of informative variables, optimization of excitation signals, and reliable identification under noisy, incomplete, and nonstationary conditions. The expected outcome is a framework for adaptive soft sensors that enhance prediction accuracy, cost efficiency, and long-term applicability in industrial environments with minimal human intervention.
|
