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

Area of Science:

• Engineering
• Control Systems
• Data Science

Background:

• Publicly available experimental datasets for nonlinear processes are scarce.
• Validating adaptive control, data-driven monitoring, and system identification requires real-world data.
• Existing benchmarks often lack the complexity of real industrial systems.

Purpose of the Study:

• To provide a comprehensive, high-quality dataset from a nonlinear process for control engineering applications.
• To support the development and validation of advanced control strategies and data-driven methods.
• To establish a benchmark for control-oriented system identification under realistic conditions.

Main Methods:

• Collected 25 long-duration time-series datasets (over 760,000 samples) from a laboratory two-tank water pumping plant.
• Conducted experiments under closed-loop adaptive control (1-second sampling period, ~8.5 hours each).
• Recorded operational data including timestamps, valve positions, tank levels, control signals, errors, adaptive PID gains, and estimated model parameters using recursive least squares.

Main Results:

• Ensured data integrity through exploratory analysis, controller behavior evaluation, and anomaly detection.
• Captured significant nonlinearities like turbulence and actuator saturation.
• Datasets cover steady-state conditions and disturbance scenarios with abrupt outflow variations.

Conclusions:

• The released dataset is a valuable resource for control engineering tasks, including adaptive control evaluation and disturbance rejection analysis.
• It serves as a robust testbed for control-oriented identification under realistic closed-loop conditions.
• Facilitates advancements in data-driven monitoring and control strategy validation.