Soft Disagreement-Based Adaptive Uncertainty Regulation for Fuzzy Servo Control

This paper proposes a supervisory soft disagreement framework for adaptive uncertainty regulation in Interval Type-2 (IT2) fuzzy servo control and validates its performance through embedded implementation on an industrial servo platform. The proposed framework introduces a supervisory learning layer that combines supervised classification and unsupervised fuzzy clustering to characterize servo operating conditions using experimentally extracted performance indicators, including rise time, settling time, overshoot, steady-state error, Integral Absolute Error (IAE), control-effort energy, tracking-error standard deviation, and maximum control effort. Operating condition confidence is quantified by measuring the soft disagreement between the posterior class probabilities of a Support Vector Machine (SVM) classifier and the normalized membership degrees of a Fuzzy C-Means (FCM) clustering algorithm using the Bhattacharyya coefficient. The resulting disagreement index adaptively regulates the Footprint of Uncertainty (FOU) of the antecedent membership functions in IT2 fuzzy controller. A closed-form Uncertainty Avoider Defuzzification (UAD) strategy enables computationally efficient uncertainty-aware type reduction for real-time embedded implementation without iterative procedures. The framework was trained using experimental data collected from a Delta ASDA-B2 400 W industrial servo drive under diverse operating conditions. The complete controller was implemented on a Raspberry Pi and experimentally compared with conventional Proportional–Integral–Derivative (PID), Type-1, and fixed-FOU IT2 fuzzy controllers. Experimental results show that the proposed controller achieved an average IAE of 1.08, representing improvements of 55.6% and 27.5% over the PID and fixed-FOU IT2 controllers, respectively. Overshoot was reduced to 2.2% and settling time to 0.24 s, while the supervisory computation required only 4.55 ms, confirming real-time feasibility. The scientific significance of this work lies in introducing a new disagreement-driven supervisory paradigm that links probabilistic machine learning confidence with adaptive fuzzy uncertainty regulation. By establishing a principled connection among supervised learning, unsupervised learning, and Interval Type-2 fuzzy control, the proposed framework provides a general foundation for confidence-aware adaptive uncertainty management in intelligent control systems operating under uncertain and time-varying conditions.