Cyber-Physical System-Based Adaptive Control For Liquid Filling-Capping Processes In Industrial Environments

Abstract

The increasing demand for intelligent automation in industrial environments requires control systems capable of adapting to dynamic operating conditions. Conventional liquid filling–capping systems in industrial automation commonly depend on fixed-rule or traditional control mechanisms, which struggle to handle dynamic operating conditions such as flow variations, bottle inconsistencies, and process disturbances. These limitations reduce filling accuracy, increase spillage, and affect operational efficiency. To overcome these issues, this research proposes a Cyber-Physical System (CPS)-based adaptive control framework integrated with a Bat Optimization-tuned Dueling Double Deep Q-Network (BO-D3QN). The research gap lies in the lack of intelligent adaptive control methods capable of real-time learning and optimization in liquid filling–capping processes. Real-time sensor data including liquid level, bottle position, and flow characteristics are collected as the dataset for system state monitoring. During preprocessing, sensor noise and inconsistent readings are filtered and normalized to improve data reliability. Relevant operational features such as filling level, flow rate, and bottle alignment are extracted for decision-making. The purpose of the BO-D3QN-based adaptive control framework is to improve convergence stability, decision-making accuracy, and real-time control performance in liquid filling–capping systems under dynamic industrial conditions. The model is implemented using reinforcement learning techniques in the Python environment, where the Bat Optimization algorithm is used to tune critical hyperparameters for enhanced adaptive control efficiency. Experimental results show improved performance with 0.98 accuracy, while reducing spillage and idle time. The proposed method provides an efficient and scalable solution for smart industrial automation.