Multi-scenario datasets for system identification and adaptive control of a nonlinear two tank system

Experimental datasets are essential for validating adaptive control strategies, data-driven monitoring approaches, and control-oriented system identification methods. Despite their importance, publicly available benchmarks based on real nonlinear processes remain limited. This work presents a collection of 25 long-duration time-series datasets totaling over 760,000 samples, obtained from a laboratory two tank water pumping plant. Experiments were conducted under closed-loop adaptive control with a 1 second sampling period, where each execution spans approximately 8.5 hours. The files capture diverse operational conditions, including steady-state baselines and disturbance scenarios with abrupt outflow variations. Each dataset contains timestamps, valve positions, tank levels, control signals, tracking errors, adaptive PID gains tuned via Dahlin's synthesis, and real-time model parameters estimated using recursive least squares. Data quality assessment confirms the integrity of the collection through exploratory analysis, controller behavior evaluation across multiple regimes, and a baseline anomaly detection case study. Capturing key nonlinearities such as turbulence and actuator saturation, this resource primarily supports control engineering tasks including adaptive control evaluation, disturbance rejection analysis, and data-driven monitoring, while also providing a testbed for control-oriented identification under realistic closed-loop conditions.