Greg McMillan and Stan Weiner bring their wits and more than 66 years of process control experience to bear on your questions, comments, and problems.
Write to them at [email protected].
Greg: Many engineering departments and plants are down to one process modeling and control engineer. Promotion to managerial positions and attrition via lures to retirement have made these engineers an endangered species.
Stan: One solution to the problem is to get rid of the last one, so the company doesnt know what its missing.
Greg: After oneness is nothingness. Is there a metaphysical aspect to oneness?
Stan: Oneness with the role, goal and whole can provide the awareness for an engineer and an organization to move to a higher level.
Greg: Were always going to have some process, configuration and automation design and support engineers, so what if we could provide interactive learning tools for them to develop the understanding how the integration of dynamics and control in the whole design can help them to meet the goal of better plant performance? For example, in neutralization systems, the time lags and delays from equipment, mixing and piping, and the sensitivity of the control valve and process as seen in the titration curve are crucial to meeting opportunities for concentration control afforded by the pH measurement.
Stan: Our twoness in Control Talk is an example of this oneness, but what can be done to make the experience more interactive?
Greg: A series of web seminars sponsored by Control is planned for this fall to debut the sections of a new book with interactive software to take engineers with any degree of experience and move them a higher awareness and oneness. Now is the time.
Stan: The moment is the moment.
Greg: Most momentous. Being in the present is essential to synchronicity and success. Ultimately the performance depends upon awareness, space and steps in the present. Most of my creative ideas originate while jogging.
Stan: Oneness and attention to the present are also keys to the performance of the automation system. For example, the role of the control valve is to recognize its present position compared to the command position and immediately move. If a positioner isnt used, theres no awareness of the present location of the trim or ability to work through the resolution limit from friction and dead band from gaps in connections and linkages. Oneness with the whole translates to the valve actually changing the flow to a degree consistent with the control system requirements, and thus, gets into these issues of what actually is happening in the trim and its installed flow characteristic. While a linear installed characteristic is obviously advantageous for a flow loop, an equal percentage characteristic offers compensation for the process gain of many temperature and composition control loops. Rangeability statements for valves that focus on uniformity of the inherent characteristic or slope fail to recognize the oneness with the increase in pressure drop and stick-slip near the seat and gain compensation, leading to erroneous conclusions that a linear valve has greater rangeability. The emphasis on the present is seen here in that omission of reset action because it would lead to overshoot from too much integration of the past. Positioner control algorithms rely on proportional and rate action where the present rules. The rate action provides some anticipatory correction but with no presupposition of the future.
Greg: There are lessons for PID control of temperature, level and pressure. The PID tuning settings must reflect the process dynamics; e.g., the reset time should be equal to the dominant time constant in the loop (hopefully in the process and not in the measurement). For integrating and runaway processes where there is no apparent self-regulation, the controller gain must be large enough to take immediate action, and rate should be used compensate for any time lag, otherwise the process will continue on its path of divergence from the set point. Processes with a process time constant much larger than the loop dead time look like integrators in the control region, and are called near integrators. Too small of a controller gain creates an adverse effect from reset action, and can lead to these integrating and runaway processes hitting alarm and trip limits. Some of these loops could benefit from rate action that uses the slope of the diverging signal to provide compensation of lags and preemptive action. Rate settings that compensate for thermowell and heat transfer surface lags are essential for exothermic reactors.
Stan: Sometimes in the overzealous explanation of the benefits of models for inferential measurements (soft sensors), feed-forward and model-based control, academics forget that all models have missing parts and will have an error that increases with time because of presupposition of the effects of the past on the present and its assumed relationships and parameter values. The aging of equipment, catalyst and field devices and valves, and the integration of inputs inherent in first principle models and batch operations translates to increasing errors with time. All models need feedback correction. Control systems that use these models must be aware of the present, and decide that no move is the best move, or take a step in the right direction.
Greg: Model Predictive Control (MPC) learns from its past moves to provide a vision of the future, but relies upon feedback action from the present measurement of the controlled variable to shift its trajectory. MPC would not be used in industry if there were no feedback correction. MPC inherently understands its oneness with the whole plant and time frame by incorporating the dynamic response of multiple manipulated and disturbance variables and constraints, as well as economic objectives. It also takes a longer term view of the error and the correction time frame, important to avoiding over-response to short-term or abrupt changes in the controlled variable or too large a change in the manipulated variable via a move size limit. Consequently, an MPC has inherent advantages over PID for addressing interactions, noisy or stepped measurement responses, and optimization opportunities. Whats missing from MPC is rate action for integrating and runaway processes.
We conclude with the Top 10 things college students dont learn in their course on process control.
Missing in Process Control Courses...
(10) Control valves with stick-slip and dead band
(9) Measurements with repeatability errors and stepped responses
(8) Volumes with mixing lags dependent upon agitation and geometry
(7) Volumes with transportation delays dependent on flow rate
(6) Control action (direct and reverse) and valve action (inc-open and inc-close)
(5) Control algorithms for anti-reset windup, output limits, and override control
(4) Control structure (e.g. gain on error and rate on process variable)
(3) Industry standards for function blocks and communication
(2) Control magazine
(1) My books (except for Washington University in St. Louis).
This Month's Puzzler
Ladies and Gentlemen, Start Your Acronyms
When does a PID perform better than a MPC?
Sen an e-mail with your answer to the Puzzler, CONTROL questions, or comments to [email protected]