I am really enjoying my exploration of some of the adventures possible after we manage to define “Measurement Risk”!

First, let’s start with an attempt to define “Measurement Risk”.

Measurement risk is the probability that making more measurements in any process instance would then cause us to change our mind about the quality of our process output.

I am on a “team”[1]?that?formed around an idea: It’s time for measurement people to sponsor the unification of our view of Measurement Risk and a more global Process Risk viewpoint that process operators already have. To do that, let’s acknowledge that finally we can lay our hands on the tools necessary to create an online, moment-by-moment Measurement Risk metric. Having done that, we will then share this component with the process owners and any other production functions. The result will be a much more comprehensive Process Risk estimator that everyone can share and from which all can profit.?

Among the many challenges of the environment within which we measure processes is the fact that there are moments in which Measurement Risk may appear so low as to be “trivial”. This holds only as long as 1) the relationship between measurement capability and process is “robust” and 2) the process that the measurement system monitors remains in the exact center of its range. In the backs of our minds, we should remember that even when it is low, this particular Measurement Risk metric cannot ever go all the way down to zero, because Measurement Uncertainty is a practical barrier.

We can see another challenge that occurs way over on the other end of the Measurement Risk spectrum. Experience tells us that these Risks grow exponentially whenever our process approaches the region around any process Limit, no matter who wrote the Limit, or why. Typically, process owner/operators have placed these limits to protect themselves from having to listen to irate customers.?

Regardless of who wrote these limits, or why they exist, global process risks also increase very rapidly in the neighborhood of their limits. And that is exactly my point:?Small or large,?Measurement Risks are Process Risks.?Once we accept this fact, then we can see clearly that it’s time for metrologists to hop on this bus!

A deeper challenge is that we may track a process from one end of this Risk spectrum to the other and never be positive about why it has varied, or exactly how we wound up at any particular risk level. There is no risk if we know exactly where the process is going next and why. In the absence of that exact knowledge which comes from Causality, we have, over centuries, slowly developed statistics.

Traditional statistical methods that we have employed for riding herd over a process and its variation often involve collecting items such as the process?mean, or its?standard deviation,?s?. It is no secret that these methods respond rather slowly to the very rapid, accelerating changes that we see in Risk in the region of a Process Limit. At times, we may even value that characteristic.??Using a mean value is better than using a single instantaneous value if and when we need its extra reliability and stability. Historically, we willingly sacrificed the value of having more exact information about the current process location in order to gain the strength we get in that greater assurance, reliability and stability.?Yet the current, moment-by-moment process location has always been crucial information for its owners!?This conflict has persisted for centuries and will continue because it isn’t soluble by any method that we can see around us.?

What metrologists?can?do right now is to quantify Measurement Risk in terms that process owners can readily understand and use to enhance and improve management of their current process risk. This requires metrologists to force their methods in two separate directions. Down onto the shop floor where the process runs, and away from the relatively easy perspective of the retrospective viewpoint.

Directions…

Down onto the shop floor. Away from proprietary platforms, away from over-technical, forbiddingly difficult methods like The GUM, and toward crowd sourcing, transparency, and shared methods[2].?Away from the “retrospective”?[3]. Retrospectivity carries the cultural cost that results from thinking that our primary function as Metrologists or calibrators, is to confine ourselves strictly to revealing more about possible past events. That is the only thing that calibrations are capable of doing for process owner/operators. As long as habit and culture dictates that we use tools developed and used within this very conservative framework, we will continue to define the very narrow limits to our usefulness to our process owning customers.

This particular Measurement Risk definition allows for only two outcomes for process owners over time. Either subsequent measurements would make us change our mind about our product quality or they would not. Despite its appearance, this definition isn’t theoretical.?That is because since measurement uncertainty always exists, the chance of changing our mind as a result of it must also always exist. The quality of the match between the process and the system that we install to measure/control it also has a direct effect on this probability. These conditions remain in effect even during moments when Measurement Risk appears to be at a theoretical minimum. In this moment we might even label this Risk as trivial.?Measurement quality must always pose a risk to its process. This is the crucial fact which metrologists and calibrators should hold as their main responsibility to the consumers of their measurement skills and experience.

To close, in addition to my team, I owe a debt to Cassie Kozyrkov. While holding down a day job at Google, she is acting as the midwife to a practical Decision Science which is appearing right before our eyes. Following Cassie over in the “ML (Machine Learning)/AI” space is where I got the idea of linking measurement data as the trigger that causes us to then change our minds about anything. Decision Science is a path to exploring the circumstances that accompany this very fundamental and universal action, which so far as we know, only humans can do. OMG! Whenever we change our minds, we are displaying emergent behavior!

Thank you once again for spending time with me in this fascinating neighborhood.

[1]?The other members are the esteemed duo of Greg Cenker, and Henry Zumbrun!

[2]?Please check out a perfect example of this direction: the CUBYT project which is approaching 800,000 publicly shared calibration data sheets. Or, in contrast with The GUM, please check out a page or two of “M3003” from United Kingdom Accreditation Service (UKAS). It’s written to improve upon The GUM for both beginners and experts. And it does!

[3]?When we return any calibrated equipment to its owner, we accept payment to deliver two kinds of news to them: “bad news” or “no news”. If the process owner really depends on this equipment for controlling product quality, they will have already heard other evidence from their customers that we were going to find it out of tolerance (OOT). Otherwise, our “no news” comes at the hard cost of stopping production while we calibrate their stuff and tell them something they also already knew: they haven’t had any calibration problems in the last 6-12 months.