In the ISA Automation Week Mentor Program I am providing guidance for extremely talented individuals from Argentina, Brazil, Malaysia, Mexico, Saudi Arabia, and the USA. We will be sharing a question and the answers each week. If you would like to provide additional answers, please send them to Susan Colwell at ISA.
The third question from Hector Torres in Mexico is:
How do you know when a feed-forward signal is needed? What characteristics should a feed-forward signal have? How does the lag and dead time get taken into account?
- If on a trend recording a change in an input flow or speed (load) associated with a primary process loop causes a significant change in the primary process variable, there is an opportunity for feedforward if the primary loop manipulates a secondary loop flow or speed. If the primary loop manipulates a valve, the feedforward gain is generally too nonlinear unless there is the valve has linear trim and a constant pressure drop. If the feedback time delay (primary loop deadtime) minus the feedforward time delay is positive, a feedforward correction arrives early possibly causing inverse response. The primary feedback time delay can be observed by momentarily putting the loop in manual and making a step change in output and then immediately putting the primary controller back in auto and noting the time delay till a change in the primary process variable. The feedforward time delay can be observed as the time delay between the start of the load change and the start of the process variable change with the primary loop in manual. To ensure the correction arrives in the process at the same time as the load change, the feedforward signal should be passed through a deadtime block whose deadtime is the difference between feedback and feedforward time delays for dynamic compensation of the feedforward for deadtime. If the difference is negative (feedback time delay smaller than feedforward delay), the feedforward arrives late and dynamic compensation is not possible to make it arrive sooner. For a late feedforward, the feedforward gain should be reduced by the ratio of this negative difference to the feedback loop deadtime. As the difference approaches the loop deadtime, the feedforward gain approaches zero.
- The computation of lead-lag dynamic compensation is more complicated than realized. Most of the control literature incorrectly shows the measured disturbance entering downstream of the process directly into the measured primary process variable. Since most load disturbances enter as a process, the computations for dynamic compensation by lead-lag is more complicated. I would first get the deadtime dynamic compensation right. If the response to a load change is in the same direction without the feedforward, the feedback lag is larger than the feedforward lag. A lead-lag can be added to reduce the deviation for the load change. The lead should be gradually increased. A lag should be first added that is 1/10 of the lead to smooth out noise that is amplified by the lead. Pages 222 – 232 in Shinskey’s Process Control Systems 4th edition offers details on the dynamic compensation of feedforward signals for load disturbances.
- The feedforward gain is the ratio of the change in controller output (without feedforward) to the change in the feedforward signal. The ratio calculation should be in the engineering units of the feedforward and controller output. The feedforward signal scale should be set to match the controller output scale. The operator should be allowed to enter a ratio setpoint that becomes the feedforward gain. The operator should see the actual ratio as well. The ratio setpoint should be adjusted to eliminate any persistent difference between the actual ratio and the ratio setpoint.
- Contrary to what is often portrayed in the control literature, a feedforward multiplier is often more of a problem than a solution because of scaling and measurement errors and the nonlinearity introduced. The possible exception is the ratio control of inline plug flow systems (e.g. extruders) and sheet lines with high accuracy and rangeability flow and/or speed measurements. Here the nonlinearity of the feedforward multiplier compensates for the nonlinearity of the process’s composition and temperature response. For most other systems, a feedforward summer is more forgiving, compensates for bias errors in measurements, and keeps the controller gain more constant. The controller gain is proportional to the ratio of the process time constant to the open loop gain. Most people don’t realize that in the control of temperature and composition in volumes with some degree of mixing, the process time constant and the open loop gain are both inversely proportional to total flow. A feedforward summer simply biases the computed feedforward (feedforward signal after dynamic compensation and multiplication by the feedforward gain) so that the process gain retains the inverse relationship with flow.
For more information check out the March/April 2011 InTech article “Feedforward enables flexible sustainable manufacturing”
Demo-Seminar “Feedforward Control” (Deminar#11)
Post “Flexible Manufacturing”
Post “Smart Adaptive Feedforward”
Post “Universal Concept – Bias”