Batch vs. Continuous Control and Optimization – Part 2

Batch and continuous temperature loops can have vastly different responses. Fortunately, temperature loops have a characteristic early in the response that enables a fast identification of the process dynamics. The resulting models can be used for process control improvement, tuning, and rapid deployment of models for plantwide simulations.

Temperature is the most important common measurement in batch and continuous processes. Temperature provides an inferential measurement of boiling mixtures and sets reaction, crystallization, drying and biological product formation rates.

Temperature can exhibit 3 classes of open loop responses; a self-regulating, integrating, or runaway (slides 3-5 in Rapid-Identification-of-Process Dynamics). The open loop response occurs when the controller is put in manual and a step change in the controller output is made. The self-regulating response occurs in continuous processes. The open loop response decelerates after the inflection point and lines out a new steady state if there are no load disturbances. The integrating response occurs in batch processes. The open loop process response continually ramps until a physical limit is reached (no steady state). The runaway response occurs in highly exothermic reactors particularly polymerization reactors where the increase in reaction rate with temperature and concentration causes a heat release that exceeds the cooling capability typically for large disturbances or abnormal operating conditions. The open loop process will accelerate until a physical limit is reached (no steady state). The acceleration may not be noticeable for small increases in temperature. The process is open loop unstable. For safety, the temperature controller may need to stay in automatic precluding an open loop test.

Fortunately, the temperature response looks like ramp for all of the different types of processes in the beginning of the response after the deadtime. A self-regulating process ramps until the response decelerates. A runaway process appears to ramp until the process noticeably accelerates. The maximum ramp rate divided by the change in controller output is the integrating process gain that can be used for process analysis, tuning, and simulation. Typically the maximum ramp rate after just 3 deadtimes is sufficient unless there is a large secondary time constant compared to the deadtime or an inverse response where the initial response is in the opposite direction of the final response. The method can monitor the process response for the maximum ramp rate for as many deadtimes deemed necessary to avoid problems with non-deal first order plus deadtime responses. The test can be done for a setpoint change with the controller in automatic if the PID gain is larger than 2 and the structure is “PI on Error D on PV” or “PID on Error” so there is a significant step change in controller output. Derivative action helps compensate for a secondary time constant. Runaway processes typically have a high PID gain and a test in automatic is the generally the best option.

Critical for success is the use of the technique described in the March 4 post A Calculation so Powerful Yet so Simple.  The new process variable (PV) in this case the current temperature measurement is passed though a deadtime block to create an old PV. The deadtime block deadtime is set equal to the observed total loop deadtime. The old PV subtracted from the new PV is a delta PV that divided by the deadtime block time interval (loop deadtime) is the rate of change. The maximum rate of change in the correct direction divided by the change in controller output is the maximum integrating process gain. The calculation with the deadtime block provides a continuous train of the ramp rates updated each module execution. Any PV filtering should be done on the new PV before the calculation (slide 6). The use of the loop deadtime as the time interval improves the signal to noise ratio. The maximum controller gain for maximum disturbance rejection is simply about 40% of the inverse of the product of the integrating process gain and loop deadtime. The methodology was originally thought to be only applicable to near-integrating processes, which are self-regulating processes with an open loop time constant so much larger than the loop deadtime that the process response looked like an integrator. Surprisingly for deadtime dominant self-regulating processes, the results also provide the upper limit to the controller gain as being simple 40% of the inverse of the open loop gain as developed in the attached document Simple-Universal-Method-for-Computing-PID-Controller-Gain-Rev1. Finally, the methodology is suitable for runaway processes with tests done in automatic.

The calculation can be setup to start whenever there is a significant step change in controller output whether due to a manual output change or a setpoint change. This simple, fast, and nonintrusive approach can be used to provide experimental or hybrid models for a plantwide simulation (slides 7 – 9).