Online intelligent knowledge extraction from real-world nonstationary data streams presents a multiobjective optimization challenge. Here, we characterize the learning process on a trajectory of global optimality to simultaneously satisfy six high-profile objectives: 1) optimum generalization for the best bias-variance tradeoff; 2) compactness of knowledgebase; 3) memory retention and stability-plasticity balance; 4) universality and full autonomy; 5) robustness against outliers, noise, and model uncertainty; and 6) active concept drift detection and adaptation. We propose a flexible Takagi-Sugeno (TS) fuzzy system, named UFAREX, that self-constructs and self-guards from scratch in a non-iterative sample-wise training scheme without storing data. Through quantification of various uncertainties, an adaptive prediction interval (API) is sequentially learned for each local dynamism to automatically capture the most accurate compact representation with a 95% confidence. This leads to the best linear unbiased estimation (BLUE) of local trends. To avoid catastrophic forgetting, API collaborates with trapezoidal membership functions (TMFs) to expand local boundaries with maximum plasticity and without distortive extrapolation. As a robust detection mechanism, API also pinpoints regions in conflict (RIC) where concept drifts and outliers are actively expressed w.r.t. time of occurrence, location, type, and severity. This establishes a single-sample online active concept drift management with zero buffer latency for regression applications. No heuristic forgetting, pruning, splitting, merging, and weighting mechanisms are exercised to prevent human intervention and render universality. UFAREX was comparatively tested on four real-world benchmarks. It stands out as an autonomous system geared for adaptive modeling, time-series forecasting, anomaly monitoring, and robust fault detection and diagnosis.