In part 5 we finish with a list of my foremost best practices. These practices build on the essential concepts given in Part 3. These practices offer simple fixes in the automation system design. Major improvements in the mechanical design are also introduced. The concepts and practices will be featured in a future article coauthored with Michel Ruel and Jacques Smuts "What Every Engineer Needs to Know about Process Control"
Best Practices for Process Control
•17. Minimize total loop dead time. The minimum peak and the minimum integrated error for disturbances is proportional to dead time and dead time squared, respectively. If you use setpoint feed forward or bang-bang logic to achieve the fastest possible setpoint response, the rise time (time to reach setpoint) is inversely proportional to the dead time. These relationships are discussed in the InTech Jan-Feb 2012 article and online appendices "PID tuning rules". The ultimate period is proportional to the dead time (natural frequency inversely proportional to dead time). Since loops can only attenuate oscillatory disturbances whose period is significantly larger than the ultimate period per remnants of what we learned from Bode Plots as depicted on slide 8 in the ISA-Edmonton-Short-Course-Effective-Use-Measurements-Valves-PID-Controllers.pdf, decreasing the total loop dead time allows us to handle faster oscillatory disturbances. The higher priority sources of dead time are the ones that are quick and inexpensive to fix that total 10% or more of the loop dead time. For flow and pressure control, the priority would be to increase digital positioner gain and to decrease variable speed dead band, transmitter damping, PID module execution time and wireless update time. For these and all loops, the backlash and stiction of valves should be minimized since these add dead time proportional to the dead band or resolution divided by the rate of change of the controller output. For temperature control the priority would be to decrease the measurement lag by improving the thermowell design (higher conductivity and lower mass) and fit of sensor (grounded and spring loaded sensor and smaller air gap within thermowell). Next priority would be to increase fluid velocity to keep sensor clean and fast by an increase in the heat transfer coefficient. Similarly for pH electrodes, higher fluid velocity will keep sensor clean and fast by an increase in the mass transfer coefficient. Next priority for temperature and composition sensors is to minimize plug flow residence time. Maximizing fluid velocity and minimizing transportation and injection delays requires working with process and mechanical design. See the July 27, 2010 ISA Interchange post "Tip #70 - Minimize Dead time" for more info.
•18. Eliminate oscillatory disturbances. Oscillatory disturbances are the most disruptive. The oscillations are amplified when the period is near the natural period of downstream loops. Frequent sources of oscillatory disturbances are loops with poorly tuned controllers, high process nonlinearity (e.g. pH), and limit cycles from dead band and resolution limits. There are limit cycles everywhere since the best valve resolution is 0.1%. It is just that the amplitude is washed out by volumes, screened out by data compression, and lost in noise or disturbances. There are exceptions to almost every rule. If you cannot eliminate an oscillation and there is a significant residence time between the oscillation and downstream loops, the strategy would be to make the oscillation faster. For example, a well proven strategy that saves millions of dollars in capital costs is to use an inline pH control system instead of a well-mixed vessel upstream of a large residence time. For strong acids and bases, the pH controller controls the average of the oscillations on the incredibly steep portion of the titration curve and the downstream volume smooths out the oscillations to the point where they are negligible. Signal characterization in the inline system to translate pH to reagent demand (x axis of the titration curve) can reduce the amplitude but not eliminate oscillations at the static mixer for control between 6 and 8 pH of a true strong acid and strong base with negligible carbon dioxide absorption. A less extensive example is where process sensitivity is high and the offset from a valve resolution limit is excessive. Using fast integral action within the positioner will cause an average position that reduces the offset in the process variable. Note that this will wear out the control valve packing so this is a last resort used only if the process gain multiplied by the minimum possible resolution limit causes a larger than permissible offset of the process variable.
•19. Slow down load changes. The step change seen in control text books is the most disruptive load change. Major sources of step disturbances are manual operator actions, sequences, on-off control, and safety instrumented systems (SIS). The culprit is normally an on-off valve. The best thing you can do is to eliminate manual actions and add a PID controller with throttling valves to eliminate sequences and switches actuating on-off valves. Fed-Batch is more kind to other users of the streams that traditional batch. Basic regulatory loop performance should be improved for many reasons to prevent activation of the SIS. For more info see the August 10 2012 ISA Interchange Post "Tip #69 - Add Control Loops to Eliminate Manual Actions and Sequences"
•20. Use the near integrator approximation for a loop with a large process time constant. The identification of the ramp rate rather than the time to steady state enables much shorter tests that are less vulnerable to load disturbances. The integrator tuning rules also provide a reset time that is much faster than the self-regulating process lambda tuning rules, dramatically reducing the integrated error for load disturbances. Also the tuning simplifies to using a minimum lambda equal to the dead time for the tightest control. This approach is especially effective for temperature and composition control of liquid reactors and columns. The near integrator gain is the ramp rate in %/sec divided by the % change in the manipulated variable (e.g. % change in flow). The near integrator gain is the open loop gain (%/%) divided by the large process time constant (sec). The near integrator approximation is also used for model predictive control to shortening the time horizon, improving the resolution of the other models since the number of data points over the horizon is normally fixed.
•21. Maximize the process time constant downstream of load disturbances. This time constant slows down the excursion from a load change and enables a higher PID gain. If the process time constant is upstream of the disturbance, it slows down the correction but not the appearance of the disturbance and is consequentially less beneficial. Control theory courses and publications often show disturbances on the output of the process leading to incorrect conclusions on the value of the process time constant. Fortunately most load disturbances occur at the input of the process. For vessel composition control, the process time constant is the residence time minus the mixing dead time. The mixing dead time is 1/2 turnover time for well-designed vessels with baffles and axial agitation. For more information on the effect of process and mechanical design on process time constants and gains, see January 11, 2013 ISA Interchange Post "Tip #79 - Understand How Equipment and Operating Conditions Affect Process Dynamics" in the ISA 2012 book 101 Tips for a Successful Automation Career.
•22. Use external reset feedback to prevent oscillations from slow secondary loops and valves, kill limit cycles, and enable directional move suppression. The positive feedback implementation of integral action offers the use of external reset feedback (ERF) to prevent a PID controller output from changing when a manipulated variable (e.g. secondary loop or final control element) is slow or not responding. ERF can prevent the burst oscillations from violation of the cascade rule that requires the secondary loop to be 4 times faster than primary loop. ERF will stop integral action when a valve is not responding due to backlash or stiction (fast readback of actual valve position is needed). Since the PID output will not change faster than the PV of an analog output (AO) or flow loop can respond, Up and Down setpoint rate limits can be added to the AO block or flow loop PID to provide directional move suppression that will prevent unnecessary crossings of a split range point or enable a slow approach to optimum and fast getaway for abnormal operation to prevent running out of valve when a valve position controller is used for optimization as described in the December 2012 Control article "Don't Over Look PID for APC". Directional move suppression can also help prevent compressor damage by a fast opening and slow closing surge valve and minimize reagent use by a slow approach to an environmental pH limit and a fast getaway to prevent a violation of the limit. Finally, ERF enables a PID to stop its integral action between analyzer updates as described in the September 28, 2012 ISA Interchange Post "Tip #100 - Use an Enhanced PID for At-Line and Off-Line Analyzers"
To avoid saturation, I will stop here and leave the rest for the future article. I hope the suspense is not too much. Next week we move on to the remaining questions from Brian Hrankowsky.
