Batch process control can seem like another world compared to continuous process control. In batch operations, process conditions are constantly changing and control loops are going in and out of service. PID control may take a back seat to sequential scheduling of manipulated flows. Here we look at why batch processes are so different and challenging and what can we do wherever possible to apply the power of PID control and Model Predictive Control (MPC). The ultimate objective here is to take advantage of the lessons learned in both worlds and enable the process control engineer to move effectively between the two worlds.
Like a perpetual startup and shutdown. Equipment is continually coming into and out of service in batch operations. Target operating conditions must be reached. As soon as these are reached, the operation moves on to the next task possibly using different equipment and different measurements and valves as quickly and as repeatable as possible.
Every phase can be a process onto itself. A batch operation may have pressurization, heating, cooling, conversion, and stripping phases. The control system must move seamlessly from one phase to another. Controllers may have different functions and objectives and certainly different tuning settings.
Wide spectrum of product grades and formulations is common. The start of each batch offers the opportunity to make a new product or at least a different grade or formulation. This can lead to complex changes in recipes that the operator and control system must deal with. In continuous processes, transitions to a different product grade or formulation are minimized or not made because the transitions take appreciable time, require extensive process knowledge and can create off-spec product that is often difficult to recycle.
Extensive sequencing and operator involvement is required. There is usually never a dull moment with a batch. Something is always changing. There is no chance to sit back and see relatively constant valve positions like when a continuous process runs at a steady state.
Dynamic response is non-self-regulating (non-stationary and no conventional steady state). The batch response of concentration, pressure, and temperature is integrating or runaway. PID tuning rules and MPC strategies based on self-regulating processes do not generally work without some modification or translation. Integrating process models and tuning rules are needed.
Extreme rangeability of manipulated variables is often needed. The cooling rate, vent rate, and feed rates for biological and chemical reactions and crystallization exponentially change with batch time. For dissolved oxygen (DO) control the oxygen uptake rate may increase by several orders of magnitude as the batch goes from the pre-exponential growth to near the end of the exponential growth phase. For fermenters, DO control is split ranged between air sparge flow, agitator speed, and vessel pressure. For bioreactors, DO control is split ranged between air and oxygen sparge and overlay flows.
Dynamic response is nonlinear. Changes in liquid volume, heat transfer surface area and coefficient, operating conditions (e.g., concentrations, pressures, and temperatures) and in manipulated variables (e.g., split ranged variables) causes changes in integrating process gains and secondary time constants and dead times.
Dynamic response may be unidirectional. For batch processes where there is only heating or only cooling and no endothermic or exothermic reactions or changes in phase, the temperature response goes only in one direction. For batch neutralizers where there is only an acid or only a base addition and no consumption of reagent in a reaction or changes in phase, the pH response goes only in one direction. For cell and product concentration in fermenters and bioreactors, the concentration only increases assuming death rate and hydrolysis is negligible. For these processes, integral action cannot be used to control temperature or pH at a specific setpoint. A PID structure of Proportional on error and Derivative on process variable (P on E, D on PV, no I) is used.
Setpoint overshoot is problematic. The integrating or runaway response makes setpoint overshoot more likely. Tuning becomes difficult and counter intuitive in that a higher PID gain may be needed to prevent the lingering overshoot from integral action. For unidirectional response, there is no return to back to setpoint. For sensitive biological processes, overshoot of a few tenths of degree or a few hundredths of a pH is undesirable. Here a very slow approach to setpoint that eliminates overshoot is desirable because the increase in time to reach setpoint is very small compared to batch time. A two degree of freedom (2DOF) structure with beta and gamma set to zero (equivalent to an I on error, and PD on PV structure) and conservative tuning may be the best choice particularly since disturbances from cell growth and product formation are so slow.
Window of allowable PID gains exists. Oscillations develop for a PID gain that is too small as well as too large due to the non-self-regulating response. For runaway processes, the process response can accelerate to a point of no return if the PID gain becomes less than the open loop positive feedback gain. Temperature controllers on highly exothermic batch reactors have this threat to an extreme where the controller cannot be put in manual for open loop bump tests and integral action is not permitted.
Contaminants, impurities, and inhibitors are trapped in the batch. Since there is no liquid discharge flow till the batch is done, concentrations of undesirable components will build up as the batch progresses.
At-line analyzer and off-line analyzer results are often too late. Analysis results are often not available until the phase or batch is completed.
Variability is trapped in batch endpoint. There is no inherent attenuation from a continuous flow through a volume. You are stuck with a bad batch requiring possibly scrapping the whole batch unless you are blending a whole lot of parallel batch trains downstream. For bioreactors, the loss of a multimillion dollar multiday batch is a huge hit to the bottom line.
Batch process yield, production, quality, and repeatability are interrelated. There may be a tradeoff in extending a batch time to gain yield or improve quality versus losing production rate. Also the ability to improve operating conditions depends upon batch repeatability. This is why batch data analytics first tries to identify what batches differ from the average batch and why. The ability to make better decisions and improvements is often related to batch repeatability just as with any measurement used for control.
Data exclusion frequently needed for batch analysis. Since equipment and associated controls are continually going in and out of service, data and alarms must be intelligently excluded.
We can learn from batch operations how to better automate the startup and shutdown of continuous processes. We can use the tuning rules for integrating processes used extensively in batch processes and the awareness of the window of allowable PID gains to tune the PIDs for continuous concentration, temperature, pH and pressure control of vessels and columns (e.g., near integrating response). We can learn about the buildup of contaminants, impurities, and inhibitors in batch processes to prevent a similar occurrence in continuous processes with extensive recycle (e.g., snow balling effect). We can use the adaptive scheduling of tuning settings needed for batch operations in continuous operations to deal with startup, shutdown, split ranged operation, and the catalysis degradation and heat transfer surface fouling with time.
We can translate the controlled variable from batch concentration, pH or temperature to a rate of change of concentration, pH or temperature for control of the desired batch profile to provide a pseudo steady state and a bi-directional response. This enables the use of integral action and MPC offering a smoother and tighter control of the batch profile. We can use some of the strategies developed for continuous reactor control for fed-batch reactor control including the use of valve position control to maximize reactant feed rate as detailed in the ISA 2015 book Advances in Reactor Measurement and Control. It also facilitates a greater improvement by the use of an enhanced PID for the large and variable update times of at-line and off-line analyzers per the 7/06/2015 Control Talk Blog. We can also use inferential measurement techniques developed for continuous processes to provide concentration measurements between corrections by analysis results for closed loop control of batch concentration. We can head off bad batches and better develop the Projection to Latent Structure (PLS) prediction of batch end point by data analytics software to make mid batch corrections.
For a summary of the challenges and opportunities in batch process control see the Chemicals & Petrochemicals Plant Automation Congress 2015 presentation “Batch Process Control Strategy”.
