Abstract
The chemical sector is lagging behind when it comes to combining process control together with machine vision due to several sector specific reasons. Vision based process control is an approach that utilizes machine vision systems to monitor and analyze real-time visual data, enabling automated adjustments to enhance efficiency in production processes. This research investigates the potential of vision driven process control in a laboratory environment that simulates a chemical production process. On this laboratory setup it is demonstrated that vision based process control for automated anti-foaming agent is achievable, given that the dosing rate is chosen correctly. Besides proving feasibility of vision driven process control in a chemical environment, this work also investigates the operational range of the control solution via a sensitivity analysis. This research shows that model recall has a large influence on the capability of the control system to consistently remove foam formations and that in case model degradation would occur, the risk is likely that the system frequently would start dosing too late. If model recall drops too low, the system is no longer able to effectively breakdown foam formations in the setup, indicating the importance of maintaining high recall. On the other hand, model precision was shown to have less of an impact on the ability to breakdown foam formations. A decrease in model precision will result in a more aggressive dosing strategy that will frequently start dosing preemptively. This preemptive dosing can lead to the positive effect that foam can no longer be formed due to the permanent presence of anti-foaming agent in the solution. These findings offer valuable insights into the potential of applying vision based process control for specific tasks within the chemical sector.
Keywords: Vision based process control; Machine learning; Control systems; Process optimiz
