The first part of this column appeared Oct. 12. It discussed how to match the basics of process control to managerial decision-making. Part two discusses what to do if control is poor, how to match control strategy with process characteristics, and some cautions.

Basic 5: If control is poor

Consider eliminating disturbances that cause poor control of the objective, as opposed to feedback correction. Design the system to have barriers to disturbances, or buffers to temper the impact of disturbances, or control loops to temper disturbances.

Consider redesigning the process to reduce delays (deadtime). This could be changing how the organization processes information, or structures lines of reporting.

Consider redesigning the process to make the final control element have greater power to drive change. This could be changing the weighting of personal performance assessment of items on annual appraisals.

Trigger control action using SPC rules. Stop tampering.

Characterize your process:

• Quantify gain, delay, and time-constant of the process. The quantification may be intuitive, and approximate. It does not need to be exact; 
• Label the process response type (integrating, open loop stable, high order, delay dominated, unstable, etc.); 
• Identify disturbances and their mechanistic path and their dynamic influence; 
• Identify interactions of your MV to other CVs and other MVs to your CV;
• Identify constraints; and
• Separate noise from real changes.

These attributes are fundamental for defining an appropriate control strategy.

Basic 6: Use a control strategy best suited for the process

Use P-only for integrating processes.

Use PI for fast acting stable processes. Make a large initial push to cause rapid change, then back off as the CV approaches the desired value. If you don’t want to make a rapid upsetting change, push more gently initially and progressively increase intensity to make the CV move to the desired value.

Sample at appropriate intervals. Don’t take control action, then focus on other things and return to observe the process after a long time. Take frequent observations, and appropriate adjustments. An early digital control rule was to have 30 observations and control actions in an open loop transient. This made the numerical control calculations nearly the same as continuous analog control. But that frequency probably requires excessive managerial attention, and I suspect 10 observations in an open loop transient may be fully adequate.

Maybe instead of feedback control, redesign the process to remove the source of upsets or disturbances. Perhaps control the influences, not the process response to them.

Don’t use mathematical calculations. We teach control from FOPDT models process models, using empirically derived gain, deadtime and time constant. However, humans are effective with control without such models. Consider decision making in cooking, catching a ball, driving a car, or managing interpersonal relations. Precise models are not needed. An intuitive sense of the process behavior is fully adequate. Even if a model is only 70% correct, and the first control action leaves 30% error, the second correction cuts that by another 30% leaving 9% error, then about 3%, rapidly becoming unnoticeable after 10 sequential control corrections.

Consider a cascade control strategy. You don’t want to wait to see reportable incidents to adjust the energy used in influencing employee behavior or attention. So, observe leading indicators (such as near misses, or product defects or recycled parts) that would imply inadequate or excessive employee effort and adjust the influence based on the leading indicators. If the leading indicators are valid indicators, then the primary metrics will be controlled. If the primary metric is going off setpoint, then use this information to feedback new setpoint for the leading indicators (the secondary controlled variable). This is a cascade application. If composition control is the primary desire for a distillation column, we use tray temperature, a secondary variable, for control because composition analysis usually has a long delay.  In a cascade arrangement, the primary composition controller determines a setpoint for the secondary temperature controller. The secondary controller adjusts heat input to keep the T at setpoint quickly compensating for disturbances, preventing them from affecting composition.

Get your subscription to Control's tri-weekly newsletter.

When there is a leading indicator of a process response, use it to make control decisions.

Consider ratio and feedforward control. Both approaches observe an incoming disturbance and adjust the control response to compensate before the disturbance influence can be seen. Perhaps you have hired or transferred in bunch of new employees, or merged organizations. Train them on desired procedures and customs before their past undesirable patterns are expressed.

Consider override. Here one controller overrides the decision another makes for safety, constraint, or such purposes. If the primary controller is not stopped, it will continue to try to control and progressively become very aggressive. Be sure the primary controller is not going to be penalized in their annual appraisal for losing control!

Filter to temper noise.

Some human processes are compliant. Others are willful or obstinate. Some require a strong force (a high gain controller) to make them move; for others a low gain is all that is needed. Be sure you tune your controller so that it is appropriate for the process.

Know the constraints. These may be on the MV, the CV, or on AuxVs. They may be on level (value) or rates of change. Choose control action that avoids constraints.

Caution

If you are seeking to influence managers, perhaps several levels above you with much more professional experience, you cannot tell them how to do their job, especially if you are junior in rank, you don’t understand their context of the process, or constraints. As the saying goes, “If all you have is a hammer, everything looks like a nail.” They may view you as a novice who just acquired the hammer as a tool, wanting to pound it on everything.  If you wish to influence (control) their approach to managerial control, then understand how their personal human process will respond to your preference for a control action. Use the above principles to formulate your attempt to control their behavior.

Perhaps instead of providing them with the solution, be willing to reveal the generality of how a control perspective (sensor validity, process dynamics, gains, frequency and interactions) can help managers structure their control loops. Maybe lead by sharing this article with them. Then ask if they’d like to see what a control perspective might contribute to something that they are working on.

Use their language, not the control jargon (such as proportional kick, cascade, lag, deadtime, etc.). Do not use block diagrams that represent information transmission. They are accustomed to seeing process flow diagrams where connecting lines between blocks represent material flow. You may need to explain control terminology with language and figures that make sense to them. 

Acknowledgment

Many concepts presented in this article have been fine-tuned by discussions with members of the IFAC Taskforce on Control Concepts for Managerial Decision Making.  I am especially grateful to input from Danny Abramovich (Agilent Technologies, USA), Francesco Cuzzola (Dassault Systèmes, Germany), Lane Desborough (NudgeBG, USA), Beny Grosman, (Medtronic, USA),  Oleg Gusikhin (Ford Motor Co, USA), Exko Juuso (U of Oulu, Finland), Zdzisław Kowalczuk (Politechnika Gdańska, Poland), Michael Lees (Carolton and United Breweries, Australia), Iven Mareels (IBM, Australia), Bhagyesh Patil (John Deere Technology Center, India), Stefano Pickl (COMTESSA, Germany), and Tariq Samad (the Taskforce leader, U of Minnesota, USA).