Benefits of Computing and Trending a Future Process Variable

The following tip is from the ISA book by Greg McMillan and Hunter Vegas titled 101 Tips for a Successful Automation Career, inspired by the ISA Mentor Program. This is Tip #90.

101 Tips for a Successful Automation CareerA change anywhere in a control loop will circle the loop in one loop deadtime. Every part of the control loop in the field and in the control room will be affected by the change within one deadtime. Loop components in the path as the change propagates will see the response before one deadtime but the effect of the change will not be seen at the point of origin until one loop deadtime has passed.

This concept is essential because people can get confused about the effect of digital measurement delays resulting from wireless update times and about controller scan times and execution times. If tests are made by making a change in a measurement right before a wireless device reports or the controller executes a scan or algorithm, there is no apparent deadtime. What you need to realize is that a process disturbance can arrive at any time within a wireless update time, controller scan time, or module execution time. Furthermore, the response of the controller has to pass through the delays in the loop to get to the point of measurement or control, and there is no guarantee that the response will arrive at the point of measurement or control right before it is ready to update. Statistically on average, the change will occur in the middle of the digital device time interval. The equivalent deadtime is one-half of the digital device time interval plus latency; that is, the time to output a result after an input is received. The latency for most digital devices today can be taken as zero. The latency for chromatographs is the full cycle time, leading to an analyzer deadtime that is 1.5 times the cycle time. The phase shift and the ultimate period confirm these deadtime approximations. (For more on the importance of deadtime, review Tip #70.)

A change in controller output does not immediately change the controller’s process variable (PV). The fact that nothing is seen, not even a partial response, until one deadtime in the future makes anticipation of what is going to happen difficult for humans and control systems.

A deadtime compensator can show the PV response one deadtime into the future, based on a change in controller output. A model predictive controller (MPC) can show the future PV trajectory based on the past number of MPC moves (changes in MPC output) and an experimental dynamic model of the process response to a move. What is missing in both future values is the dynamic effect of disturbances and load changes. The MPC will bias the trajectory based on the error between the predicted current PV and the actual PV, but the shape of the trajectory will not change.

A future value computed from the identified ramp rate reflects the effect of controller actions, unknowns, disturbances, and loads without any preconception of the process dynamics other than loop deadtime. For abnormal operation, start-ups, nonlinearities, batch operations, and non-stationary responses, the ramp rate method offers a better view of the future than MPC trajectories.

Concept: A future PV value can help operations understand the effects of their setpoint and manual output changes and be more patient. A future PV value can help a PID controller optimize setpoint changes and deal with abnormal operations by various output tracking strategies (Tip # 91). A future PV value can also be used for batch cycle time and yield optimization (Tip #96).

Details: Compute a future value one deadtime into the future by multiplying the ramp rate identified with a deadtime block (Tip # 89) by a time interval. To see the effect of doing this one deadtime into the future, multiply the ramp rate by the deadtime. This computation can be simplified to the delta PV created by the deadtime block (Tip #89) added to the current value. To provide a projection further into the future to provide more anticipation, multiply the PV ramp rate by a time greater than the loop deadtime. For batch profile and end point control, the time interval must be large enough that noise in the batch profile slope calculation does not affect the batch optimization (Tip #96). For smart integral action (Tip #92), increase the time interval as necessary to prevent overshoot. For control loops with controller gains much higher than 1, the future is better seen in the future controller output (CO). The CO ramp rate is computed the same way. Scale limits and output limits set the minimum and maximum future PV and CO, respectively. Plot the future values and the ramp rates on the trend chart per the checklist in Appendix C to help the operators. Consult tips #92, 95, and 96 to learn how to use the future PV value in process control. For stationary continuous processes at normal operating points, a future PV value from an MPC trajectory is better because the trajectory provides more information and is largely determined by previous MPC moves.

Watch-outs: Transportation deadtime and injection deadtime in a process are inversely proportional to flow rate. For pH processes with small reagent flow rates, the injection deadtime is particularly large. Batch and start-up deadtime will increase as the equipment is filled. The deadtime will also change as process conditions change during the progress of a batch or start-up.

Exceptions: The future PV value calculation will not work for deadtime dominant processes, noisy processes, and valves with excessive deadband and stick-slip.

Insight: A future PV value can make setpoint response, reset, and feedforward smarter and help operations understand the consequences of the dynamics on current actions.

Rule of Thumb: Compute, trend, and use a future PV, and, if advantageous, a future CO value.

Look for another tip next Friday.

Print Friendly

, , , ,

Pin It on Pinterest

Share This

Share This

Share this ISA post with your friends!