HarperMetrology

05/04/2025

25/10/2024

Metrology is everywhere.

I already knew this, but it was just reinforced to me this week. As I had some unexpected time in Manchester this week, I decided to take a bit of a walk around. As I was wandering around the Piccadilly Gardens area, I saw some large statues. There was a large monument to Duke Wellington, a portrayal of Queen Victoria and the one that really caught my eye was one of James Watt. James Watt (1736 - 1819) is well known for his work during the industrial revolution for improvements to the steam engine. He is also the namesake for the unit of power, the Watt.

This is truly an example that metrology can be found anywhere.

24/10/2024

Another option for decision rules is to apply a guard band. In a recent training, I was made aware that the idea of guard banding is primarily a topic used in America, and as such others may have less knowledge on how guard-banding works.
A guard band, when applied to calibration and specifications, is a value that is subtracted from the specifications of an instrument to minimize the probability of having a false acceptance of measurement data.
An example of this would be applying 20 % guard band to a specification. If a resistor has a specification of ± 0.5 Ω, then a 20 % guard band would be equivalent to 0.1 Ω. Applying this guard band means that in order for the resistor to pass calibration, the value would need to be within ± 0.4 Ω of the nominal value.
A common approach to creating guard bands, is to use some multiple of the uncertainty of the calibration (U) as the value for the guard band. The following are some common guard band values and their associated risk values.
For the guard band, w, as set forth in ISO 14253-1:2017: w = 0.83 * U, this has an associated PFA of < 5%.
Guard band if using ILAC G8:2009: w = 1 * U, this has an associated PFA of < 2.5 %.
3 sigma guard band: w = 1.5 * U, the associated PFA is < 0.16 %.
And lastly, 6 sigma guard band: w = 3 * U, the associated PFA is < 1 ppm.
The idea is that if a measurement result falls within the desired guard band, then you know what the associated PFA is. In the images shown, the different guard band levels are represented. The example uses a measurement which has a specification of 10 ± 1. The uncertainty of the measurement is 0.2 or 5 to 1 TUR. The different lines represent the within what range a value must fall in order to pass calibration based on the different guard bands.
For each of the rules listed, there is also a chart showing how the distribution function works together with the guard band to show how much risk of PFA there is.

Interested in learning more? Don't hesitate to contact me.

23/10/2024

Today's decision rule up for discussion is the ILAC G8:2009 rule or the non-binary rule.

This rule differs from the simple acceptance rule in two significant ways:
1. This rule uses the measurement uncertainty as part of the decision making process, and
2. Instead of just two (binary) outcomes there are actually four outcomes using this rule.

The measurement uncertainty is applied by adding (or subtracting) it to the measurement result to get a band of values in which the value can lie with a confidence level of approximately 95 %. This is the checked against the limits. The following four outcomes are possible:

1. Pass - The measurement result and the entire band of values falls within the acceptable limits for the measurement.
2. Conditional Pass (Pass*) - The measurement result is within the acceptable limits, however, some of the band of values exceeds the acceptable limits for the measurement.
3. Conditional Fail (Fail*) - The measurement result exceeds the acceptable limits, however, some of the band of values is within the acceptable limits
4. Fail - The measurement result and the entire band of values exceed the acceptable limits for the measurement.

22/10/2024

The first decision rule that I would like to discuss is called the simple acceptance or binary decision rule. This rule simply states, if the measurement result is within the specified limits, the data passes. If not, the data fails. Regardless of the measurement uncertainty.

This seems pretty straightforward and easy right, not so fast. The laboratory must understand and document the risk level associated with any decision rule applied.
The ISO 17025 standard, in section 7.8.6.1, states: "When a statement of conformity to a specification or standard for test or calibration is provided, the laboratory shall document the decision rule employed, taking into account the level of risk (such as false accept and false reject and statistical assumptions) associated with the decision rule employed and apply the decision rule."

This means as a calibration laboratory, even if you apply this simple acceptance rule, you have to understand the risks involved. With the simple acceptance rule, the risk of a possible false acceptance (PFA) can be as high as 50 %. Because each measurement has a level of uncertainty, when a measurement result approaches the acceptable limit, there is an ever increasing chance that the actual value of the measurement result is beyond the limits until the value is right at the limit in which case there is a 50 % probability that the actual value is beyond the limit, which means 50 % PFA.

Do your customers understand what the risk truly is?

21/10/2024

This week, I am going to be discussing decision rules. This is a topic that creates a lot of discussion in accreditation and calibration circles. What is a decision rule, and why do we need to use them?

Let’s start off with a real life example:

On 20 October 2024, I was on a flight bound for Dublin International Airport, to perform scheduled assessments for INAB. As we approached the airport, it was obvious that things were a bit different than normal. We began our descent and to say it was a bumpy ride was an understatement. We continued toward the runway. We got to within a hundred meters or so of the ground and the pilot increased speed and pulled out of the descent. The pilot announced that the ground winds were out of the limits for the aircraft and it wasn’t safe to land. This scenario repeated two more times and we were eventually rerouted to Manchester.
This is an example of a decision rule, the captain was given information from the sensors in the cockpit. The alarms stated that it was unsafe to continue and the landing was aborted. Could the pilot have safely landed the plane? It is possible, but the risk involved was not worth trying.
A decision rule is defined according to the ISO 17025 section 3.7 as: [a] rule that describes how measurement uncertainty is accounted for when stating conformity with a specified requirement.

In my case, the measurement stated that it was unsafe to land and the pilot decided to follow that result. As calibration personnel and metrologists we have to apply decision rules to our measurement results to determine if the data is good or not.

Keep coming back for more information, or message me directly.