Change-aware online identification and adaptive PI control for robust electrical conductivity regulation in fertigation mixing systems

Electrical conductivity regulation is essential in fertigation systems, but its performance is often limited by transport delay, mixing inertia, and operating-condition-dependent dynamics that degrade fixed-gain control. This study proposes an integrated control-oriented framework for robust EC regulation by combining offline process characterization, change-aware online identification, and low-complexity adaptive PI control. Step-response tests at five peristaltic pump speeds from 50 to 600 rpm were used to identify first-order-plus-dead-time models with high fit quality, with R² values ranging from 0.980 to 0.990. The identified steady-state gain, time constant, and effective dead time varied systematically with operating point, supporting the use of the FOPDT structure as a nominal control-oriented model of the dominant EC dynamics rather than a complete physical description of the fertigation process. A fixed-gain PI controller tuned from the nominal 200 rpm model was used as a baseline. After aligning simulation and experiment to the onset of control action, both showed the same overall closed-loop trend but different transient characteristics, with rise times of 12.7 and 20.6 s, settling times of 42.5 and 23.0 s, and overshoot values of 5.6% and 1.5%, indicating that the offline model is suitable for nominal tuning but insufficient under varying conditions. To address this limitation, a change-aware recursive least squares algorithm was used to estimate time-varying parameters online and update PI gains through simplified internal model control rules. Under process-parameter variations, the proposed CA-RLS-PI controller achieved the shortest recovery time of 13.0 s and the lowest integral absolute error of 11542.55 among all compared methods. Experiments on a laboratory fertigation platform further confirmed practical feasibility: after a concentration change introduced at about 120 s, EC peaked at 2332.6 μS/cm and returned to 1526.3 μS/cm by 242.5 s, while pump speed adjusted smoothly from about 74 to 32 rpm. These results demonstrate that the proposed framework improves dynamic regulation and is practically implementable for real-time EC control in fertigation systems with time-varying dynamics.