Machine Learning-Driven Online Tuning of Discrete PID Controllers for Non-Linear Temperature and pH Processes in Chemical Reactors

Precise control of temperature and pH in continuous stirred-tank reactors (CSTRs) is challenging due to strong nonlinearities, time-varying dynamics, and process disturbances. Conventional Proportional-Integral-Derivative (PID) controllers with fixed parameters may experience degraded performance when operating conditions deviate from the nominal tuning point. This study presents and evaluates a machine learning-assisted control architecture in which Deep Reinforcement Learning (DRL) is used to dynamically tune the parameters of a discrete PID controller. The proposed framework employs the Deep Deterministic Policy Gradient (DDPG) algorithm as an online supervisory agent that generates incremental adjustments to the PID gains KpK_p, KiK_i, and KdK_d. The agent observes the control error, integral of the error, and derivative of the process variable, while the reward function penalizes tracking error, excessive control effort, and overshoot. The proposed ML-PID controller was evaluated through simulations of nonlinear temperature and pH control processes and compared with conventional PID controllers tuned using the Ziegler-Nichols method at a nominal operating point. For pH neutralization, the ML-PID reduced the settling time from 125.0 s to 72.5 s, corresponding to a 42% reduction, while reducing the ITAE metric from 1845.3 to 412.8. For temperature control under a 15% feed-concentration disturbance, the ML-PID reduced the maximum temperature deviation from 15.8% to 1.5% and reduced the settling time from 88.0 s to 45.0 s. These simulation results indicate that DDPG-based online PID tuning can improve control performance under the considered nonlinear operating conditions while preserving the familiar PID control structure. Further experimental and robustness studies are required to assess performance under sensor noise, model-plant mismatch, and real industrial operating conditions.