There is a process control way, an approach of looking at processes and determining how to control them. Managers seek to control organizational processes and outcomes, but most managers have not been trained in the methods of control. I think that control engineers and technologists could help decision makers in management improve their method.
Promote control concepts and issues, not algorithms and theory
We have control algorithms such as PID, MPC, ADRC and SMC. They take inanimate process data and make an inanimate process influence decision. Control technologists are comfortable with giving automatic decision-making capability to such, in low CV and MV situations, where the one algorithm rule works for every situation. However, I don’t think that managers will let such algorithms have autonomous control over animate processes. The concepts behind the algorithms should be useful and acceptable.
We should not be seeking to promote control theory. Instead, we should help managers understand the fundamental concepts that relate to decision-making.
Similarly, we have several model types that have proven useful in control. These include FOPDT, ARMA, ABCD and FIR. The best values for model coefficients are not important. The model functional forms are not important for those making managerial decisions in very complex nonlinear, constrained, multivariable, uncertain space. What is important is the control insight from understanding the concepts behind the models.
What control tells us
First, understand your process. Know what reacts to what, and how. Use a phenomenological view, the sequence of mechanisms, and the timing. Know what you want to achieve and how to measure it. Know what you can manipulate (change) and how you must respond. Understand the difference between controlled variables, auxiliary variables and constraints.
Control keeps things at a set point. Keeping things the same is not improvement. Good regulatory feedback control improves uniformity which can be leveraged to process or organizational improvement by minimizing losses, reducing or eliminating upsets, improving uniformity, safely operating closer to constraints, increasing flexibility, providing surer and faster transitions, reducing false actions, and operating closer to JIT and supply chain boundaries.
Rational assessment of KPIs (due to better sensing location and phenomena and filtering), provide better knowledge, which leads to better decisions. Be sure to properly assess the objective. Errors in the measurement could be from sensing the wrong thing, calibration error, etc. Or even, purposely created by an attempt to distort the situation to seemingly justify some action. If you want to do what is right, be sure the measurement provides a legitimate representation of the goal.
P-action makes immediate push to move the process back to the desired point or toward a new point. The magnitude of the push depends on the process sensitivity. Compliant processes do not need a big push. Recalcitrant processes need a significant push. Emergency action needs a big push.
I-action continually shifts the base controller push as the P-action fades as the process moves. I-action is needed to continually adjust the control action or else there will be steady state offset. The rate that I-action adjusts the push depends on the rate that the process responds.
If the controller is not in charge, the I-action will cause the manager to wind up. Be sure the controller has the needed leverage and connectivity. Don’t try to control something that is not controllable.
Desired change does not come with an initial decision action. Control requires continual guidance.
D-action shows us to not just react to the current deviation, but to use rate of change to forecast the magnitude of the deviation a bit into the future. It also shows that extreme forecasting can lead to excessive go-stop messages.
When processes have long delays or many stages are required to process the change, the controller must be patient. It must give the process time to change, or else it causes overreaction.
Place sensors, assessment, and measurements as close as feasible to the first possible indication. Eliminate delays and lags in information about the process state.
Noise in the measurement, the KPI, leads to noise in the controller action. When a process is noisy or uncertain, use SPC concepts to temper action, to wait until action is statistically confidently needed.
On-off control is often good but consider deadband around the measurement to prevent on-off chattering.
Override control reveals that some safety or emergency action will override a primary controller action, and when this happens the out-of-control controller will wind up. You need to place primary ambitions on hold when there is an override. Planning needs to shift the prior schedule, and annual performance appraisals need to recognize the override and not hold employees or units to the normal schedule. When a project is placed on hold, those who have invested their career success in the project need to be un-wound.
Cascade tells us to seek leading indicators or intermediate variables that reveal what is developing in the process. If the leading indicators reveal all is on target, don’t take control action, but if the leading indicators reveal a change, don’t wait for the change to fully develop in the process, take action now to counter that change. Use feedback control to adjust the desired value of the leading indicator.
Ratio control tells us to scale the control with the need. If something doubles, then double the control action. If hiring increases the new employee influx by 20%, expect to increase training sessions by 20%.
Feedforward has us also monitor disturbances, influences, indicators that reveal action will be needed. But unlike ratio, which reacts immediately, feedforward teaches us to wait until just before the disturbance has an impact, then jump the control action to an initial value then relax back to a steady state value based on the process sensitivity of both the disturbance and the control action.
Interactions means that control action to fix one thing can upset another. MIMO control teaches us several techniques to fix this problem. Provide decouplers, provide a one-way decoupler if one variable is relatively important, or detune the lesser important control loop so that its corrections have little impact on the more important loop. The decoupler might be just a simple gain-scaled addition to one of the controller actions.
MPC/APC tells us to forecast the possible impact of control actions into the future (constraints as well as desired outcomes) to see what action now requires in the future.
If you are uncertain about your process, take cautious action. The process behavior changes in both time and operating point. We think of key process attributes in terms of gain, time-constant, and delay. We know that appropriate control action needs to have an associated change. People mature and employees are replaced. Organizational size and procedures change. Financial and legal constraints change. Control action must change with the situation.
The process models that we use in RTO do not account for future uncertainty. Such as tax rates, labor negotiations, raw material and utility costs, inflation. They don’t need to be because they run on an hour-to-hour or shift-to-shift frequency. But managerial decisions might set a course that requires a multi-year period to come to fruition. Managers need to include such future uncertainty in finding optimal decisions. Our work with stochastic models, and concepts from weather forecasting of possible storm paths and probability of rain, could help managers see the possible future outcomes and intuitively choose the decision that minimizes the undesirable while maximizing the desirable outcomes.
Takeaway
Help your managers understand how to look at decision making. Don’t attempt to tell them what to do. Many are enjoying their new sense of authority and don’t want underlings trying to formulate action but be willing to offer insight.
Don’t explain what control suggests with Laplace transforms!
You also manage your own arena and make your own decisions. Develop your potential. Use the way of control to guide your personal decision making.
Russ Rhinehart started his career in the process industry. After 13 years and rising to engineering supervision, he transferred to a 31-year academic career. Now “retired”, he enjoys coaching professionals through books, articles, short courses, and postings on his web site www.r3eda.com.
