S. McBul, OQ Specialty Chemicals, Musqat, Oman
Part 1 of this article, “The process control journey: Primary process control—Part 1” appeared in the August issue of Hydrocarbon Processing and discussed the performance and stability of complex-wide primary (PID) control.
Variability reduction is a pre-requisite for production maximization. For example, in FIG. 7, a plant operating away from the maximum possible production constraint (red line) can lead to a loss of maximum profit potential. Operators tend to maintain a comfort zone away from constraining limits to avoid constraint violations due to variability. This is where the science of process control comes in—the art is to first reduce variability to enable process operations closer to the constraint limit.
The following will discuss and demonstrate the step-by-step process for variability reduction—as well as maximizing production via advanced process control (APC) technology—and demonstrate the value through a case study, which is related to an APC application on a naphtha splitter in an aromatics complex (FIG. 8). The steps involved in APC projects include:
Step 1: APC strategy and benefit estimation. Once a decision is made regarding an APC project, a list of strategies must be developed that can increase profits (FIG. 8 for the naphtha splitter case study)—e.g., to increase feed or reduce energy consumption up to constraint limits. For each strategy, a monetary benefit estimate must be done so that the value of each strategy is known. For a benefit estimation, a validated process simulator can perform a quick estimate of value, or a small step test run can be performed to get the actual data and process gains. FIG. 9 shows a simulatora used to test a strategy that showed an increase in top product production in the naphtha splitter, resulting in increased gasoline production. The benefit of each strategy can be predicted with greater accuracy and APC performance can be foreseen before deployment.
Step 2: Step test, inferential and control matrix development. Once the strategy is finalized, the next step is to list all related controlled, manipulated and disturbance variables—as well as applicable constraints—and develop a control matrix that maps the relation between the variables. For example (in the case study discussed here), to control the top product quality, the controlled variables would be top product quality and feed flow, the manipulated variables are the reboiler and reflux flowrates, the disturbance variables would be fuel quality and feed quality, whereas the constraints would be column flooding rates.
Once a control matrix is developed, the models must be developed by performing step tests so one manipulated variable is perturbed at a time to determine its impact on related controlled variables. This data is recorded in an APC modeling suite for automatic dynamic model determination. This leads to the development of an APC model, shown in FIG. 10.
Note: If the controlled variable is something like product quality, which is only determined using lab tests (rather than an analyzer), then it is necessary to develop a real-time estimator (or an inferential) of this lab value as it cannot be measured but must be controlled in real time. A simple regression between the controlled variable and related manipulated variable can develop the estimation. FIG. 11 shows the relationship between inferential value vs. actual value when plotted against historical data. It is necessary that any developed inferential matches with plant value accurately or the model predictive control (MPC) may not positively impact the final end results.
Step 3: APC simulation run. Once the control matrix is developed, initialization parameters and limits are set, and APC variables are tuned, the controller must then be run in simulation mode while accessing live data from all plant variables so that its response is known and the benefit outcome can be seen in a stable manner.
Step 4: APC live run and benefits achievement. Upon a successful simulation run, the final step is to go live by connecting APC with a live distributed control system (DCS). For the initial “go live,” tighter manipulated variable (MV) movement limits are kept, including small MV changes allowed per step to avoid any surprises. Once the system is performing well in “go live,” the limits are relaxed and faster MV movement tuning allows the APC to “flex its muscles.”
Without violating constraints, the APC should slowly and steadily move the process toward optimum performance and reduce overall process variability, including the variability in inferred (invisible) lab results. FIG. 12 shows the inferential of top product quality (which is measured once daily in the lab but estimated every minute by APC): a significant reduction in variability was achieved using APC. FIG. 13 shows the trend of the same top product quality lab values—interestingly, as the variability in the inferential reduced, so did the variability in lab value, proving the quality of the inferential. This is a good example of how APC reduced the variability in lab values by measuring only once per day, thereby stabilizing the process and moving it toward optimum performance. If such a reduction in variability or optimization is not seen, then re-modeling must be performed to avoid lost value. FIG. 14 shows the overall stability achieved in all parameters of the column due to the combination of value position tuning, sensor filtering, PID tuning and APC.
In this application, the cutpoints stabilization and optimization using APC resulted in $15 MM/yr–$20 MM/yr of increased production with a return on investment (ROI) of 2 wk.
Digital RTO. Although APC optimizes local units, market changes may dictate that cutpoint limits in APC need adjustment to impact benzene vs. xylene vs. gasoline production. To overcome this challenge, a digital real-time optimizer (RTO) was developed in Excel VBA and deployed. The optimizer takes plant data and product prices from a real-time database (RTDB) and complex-wide constraint across the aromatics complex to determine the cutpoints and optimization targets. A hybrid digital twin model was preferred over a rigorous chemical engineering model, as the latter can be difficult to maintain and has contributed to the limited success of RTO applications globally. The RTO was kept in an open loop, providing optimum cutpoints and APC limits guidance to the operator in real time to change APC limits. This concept is presented in FIG. 15. To improve digital RTO, steady-state detection, data reconciliation and dynamic compensation were built. It is planned to close the loop with APC once the operators gain confidence in the optimizer. HP
NOTES
a KBC’s Petro-SIM
SAQIB MCBUL is the Head of process control and digitization at OQ Specialty Chemicals’ Oman refinery and petrochemicals complex. He has served in various plants helping owner-operators in the hydrocarbon processing industry increase their bottom lines using regulatory and advanced process control, manufacturing execution and digital information systems. Mcbul is the founder of Apex Digitization.