What configuration changes to control strategies can readily take advantage of improvements in measurements and final control elements? Control strategies can inherently improve production rate by maximizing feed and efficiency by enforcing material and energy balances. In part 3 and next week in part 4 we look at control strategies for unit operations that have a major effect on product quality, production rate, and energy use. The configuration modifications can often be made and tested within a few days if a virtual plant is used to prototype and test the control strategies for the entire range of possible process conditions including abnormal operation.
The most straight forward way to maximize an exothermic reactor production rate is to have a cascade control of reactor temperature setting lead reactant feed flow. However, the composition response is put in series with thermal response. For large back mixed liquid volumes (agitated reactors) the deadtime from the time lags in series can cause an unacceptable slow response to feed and coolant disturbances by a decrease in the allowable gain and increase in reset time for the temperature controller. If the feed is much cooler than the reactor temperature, there can be an inverse response that requires further detuning of the temperature controller. However, for plug flow volumes with short residence times and consequently by necessity fast reaction times, the composition response is so fast that the effect is negligible compared to other sources of deadtime such as measurement lags. Inverse response does not have time to develop. The controller tuning is largely set by the process dynamics and the sensor location, speed, threshold sensitivity, and noise. Slide 10 in the ISA Automation Week 2011 tutorial ISA-AW-2011-Biological-and-Chemical-Reactor-Control-Opportunities.pdf shows the highest of a multitude of bed temperature measurements on a fluidized bed gas reactor is selected as the process variable of a temperature controller that sets lead gas reactant feed rate. The other gas reactant feed rates are ratioed to the lead reactant feed rate. The production rate is maximized to match the cooling rate available that is set by the number of boiler feed water (BFW) coils in service. Inline plug flow liquid reactors with small residence times (e.g. static mixers) may benefit from a similar strategy.
For a reactor that has both gas and liquid reactants, end point control can be used to provide tight material balance control as seen in slides 11 and 12 in the reactor tutorial. For a liquid product, a reactor pressure controller manipulating the gas reactant feed will inherently set just the right amount of gas reactant flow to match what is being consumed by the reaction. This strategy works for fed-batch and continuous reactors. For a gas product, a reactor level controller manipulating the liquid reactant feed will inherently set just the right amount of liquid reactant flow to match what is being consumed by the reaction. This strategy is not used for fed-batch reactors since the level is increasing.
The conversion rate in an exothermic reactor can be controlled by an inferential measurement of the heat transfer rate. Slide 26 in the reactor tutorial shows how a synchronized measurement of jacket inlet and exit temperatures and knowledge of the jacket coolant flow can be used to compute a heat transfer rate that can be integrated over the residence time of continuous reactors and the reaction cycle time of batch reactors to provide an inferential measurement of conversion and yield if there are no side reactions.
There is an optimum batch profile of key process variables (PV) such as temperature and concentration. However, the change in PV is in one direction only (usually increasing). The direct control of the PV is not feasible because feedback control is based on being able to make changes in both directions. The slope of the profile can be controlled since the slope can decrease and increase. By simply passing the PV through a deadtime block with a deadtime set equal to the process deadtime, a continuous train of slope values is generated that can be the controlled variable for PID or model predictive control (MPC). Slides 27 and 40 in the reactor tutorial show the profile control strategies for chemical and biological reactors, respectively. Slopes are maximized at the most opportune times in the batch. For polymers the polymerization rate may be steadily increased by a temperature increase until adverse reactions can occur. For bioprocesses the cell growth rate may be maximized in the first half of the batch and the product formation rate may be maximized in the second half the batch. Profile control is more efficient than switching to higher temperature or concentration setpoints because processes generally do not themselves make discrete changes in terms of what is an optimum. Even more disparate is the advantage over the scheduling of feed rates.
To summarize, fed-batch operation by feedback control of process variables as shown on slide 30 for bioreactor substrate control is more efficient than batch sequencing of flows and profile control is more efficient than setpoint switching of key process variables. The Control July 2008 article "Unlocking the Secret Profiles of Batch Reactors" reviews this opportunity. The article discusses how we are in fact flying blind in most batch processes. Often the lab results on raw material compositions are not available until after the feeds have been completed and lab results on final batch compositions are not available until after the batch has been transferred.
A more efficient batch operation can be taken as an increase in endpoint, an increase in yield, or a decrease in batch time. For many batch operations, the maximum possible concentration may not be much greater than the normal endpoint due to side reactions or product degradation or in the case of biological reactions because of cell death, product formation inhibition, and consumption of the product by cells.
For ethanol, inhibition primarily limits the maximum batch concentration. Batches are run longer than necessary by 4 to 8 hours to insure reaching the endpoint. Efficiency is thus gained by decreasing feed rate or reducing cycle time for early endpoints. Slide 28 in the reactor tutorial shows how at-line measurement of corn fermentability by a Near Infared Transmittance (NIR_T) analyzer can be used to maximize ethanol yield. The control strategy also includes an online measurement of solids via Coriolis density to control dilution rate. Finally, the control strategy shows how at-line measurement of ethanol by a High Performance Liquid Chromatograph (HPLC) analyzer in the saccharification and fermentation (SSF) batch vessels can be used to provide an average fermentation time to correct the corn feed analyzer. The optimization is done by a simple PID controller with an inferential measurement of production rate. An increase in corn fermentability detected by the feed NIR-T analyzer results in an immediate reduction in corn feed. A decrease in fermentation time detected by the HPLC analyzer results in reduction as well but delayed by the batch cycle time. For more details see the ISA Automation Week 2011 paper ISA-AW-2011-Corn-Analysis-Modeling-and-Control-for-Ethanol-Optimization.pdf
A significant portion of a continuous reactor startup time and fed-batch cycle time is the setpoint rise time (time to reach setpoint). Setpoint feedforward and a smart bang-bang type of logic for a full throttle response can be used as demonstrated in "Deminar #7 – PID Control of Integrating Processes". Equations 14a – 14c in Appendix C in the Jan-Feb 2012 InTech article "PID tuning rules" can be used to estimate the rise time to show the effect of controller gain, setpoint feedforward, and dead time. See the May 2006 Control article "Full throttle batch and startup response" for a simple way to have the fastest possible setpoint response time.
The essential part of any rate of change calculation for inferential measurements or rapid identification of dynamics in short cut tuning method in Appendix A of the InTech article is the use of a deadtime block to create a continuous delta PV train. The block deadtime parameter is set equal to the total loop deadtime. The extensive implications of this technique are noted in my modeling and control blog "A Calculation so Simple yet so Powerful."
We conclude this series in part 4 with a brief look at effective strategies for distillation, compressor, and pH control.
