This paper outlines a model predictive control (MPC) solution for a continuous reactor that reduced pH and concentration variability, and consequently reduced costs associated with raw materials usage. MPC is widely adopted in industry as an advanced method of process control and its concept is based on constraint, disturbance, controlled and manipulated variables which establish the process control strategy. The intent of MPC is to optimize costs and improve process efficiency regarding to process variability, quality control, safety and environmental risks, and raw materials consumption. In this application, the process consisted of a slurry tank in series with a reactor in a recycle loop. A gravity feed of solid raw material is charged in shots and water is continuously fed into a slurry tank, agitated and pumped to a reactor, where a base reactant is continuously fed. Process material is returned from the reactor to slurry tank creating a recycle loop. Concentration and pH are critical quality parameters and the in-process material is circulated throughout the system until the material meets specification. Once quality requirements are achieved part of the material from reactor is transferred forward to process downstream as finished goods. There is an override control for outlet flow controller and level controller in the reactor cascading to a low output selector that controls the production rate. In order to maximize production rate using less raw materials, two MPC controllers were created. The first MPC application consisted of a single controlled variable (pH), one manipulated variable (incoming base flow rate), and one disturbance variable (product outlet flow rate). The second MPC application used a single controlled variable (concentration), one manipulated variable (income water flow rate), and one disturbance variable (slurry density). Based on the historical process data, the selected variables have significant statistical correlation with the controlled variable for both applications. No constraint variables were identified in both cases due to the process characteristics. The modeling process was started by performing the plant runs. With the plant in steady state, step changes for each manipulated and disturbance variable were made causing changes in the associated controlled variable. The tests allowed the distributed control system (DCS) to calculate the gains, first order constants and dead times. This procedure was applied three times for each MPC application. Then, the model was downloaded to the DCS controllers. Confirmation runs were performed to certify the robustness of the model. Both MPC applications run independently of each other. This project accomplished the goals, improving the base reactant usage by 3.6%, reducing operating cost with low capital investment. Additionally the MPC applications have demonstrated their importance for reducing the pH and concentration variability and for maximizing process output.
展开▼