Validating Measurement with Digital and Video Records

Why it matters

Validating a video measurement is broader than seeing that the file can be played. For a time measurement you must know which event the record shows, what the time axis is based on and where the element boundary was chosen. That software shows many decimal places does not mean the image carries information at that detail. In a fixed-rate 25 frames/s record there are 0.04 s between consecutive frames. An intermediate moment that is not in the image cannot be found just by adding digits to the result. The usefulness of a measurement comes not from the size of the displayed number but from its being suitable for the decision and reproducible.

Resolution, bias and boundary uncertainty are different problems. Resolution describes the time step that can be told apart. Bias may concern the same defined event being measured systematically early or late relative to the reference. Boundary uncertainty is a lack of information about exactly in which image the event starts or ends. If the view is blocked while K01 touches the part, higher numerical precision does not solve the visibility problem. Two people marking the same blurred image on the same frame also does not prove accuracy by itself. Without a shared element definition and enough visual evidence, agreement may be a shared interpretation error.

Duration is found by subtracting the start time from the end time. So both boundaries have an effect. If the start and the end shift in the same direction, the difference may not change; if one is chosen early and the other late, the duration grows. The safe-interval calculation used here shows the worst combination of boundaries. This interval is not a statistical confidence interval and does not say that the observed difference is the exact error. Which boundary assumption was used must be written clearly in the report. The link to measurement system analysis is here: increasing the number of records does not by itself repair an unexplained boundary definition or a faulty time base.

Validation requires the acceptance condition to be set before the measurement is made. The event to be measured, the adequacy of the reference, the number of comparisons and which difference will be accepted are parts of the same plan. In a short element 0.04 s may matter, while in a long cycle the same size has a different effect on the decision. Even so, applying the same arbitrary percentage to all elements afterwards is not correct. The tolerances in this lesson are only synthetic training decisions. In a real application, the distinction the decision requires and the uncertainty of the reference used are evaluated separately; a single successful match is not approval for the whole measurement range.

The measurement plan must also make clear what repeated marking tests. The same observer reviewing the same record again gives information about the consistency of their own boundary choice. Different observers reviewing the same event help test whether the shared definition can be applied. These two studies are not the same as observing new production cycles. Measuring the same record again does not increase the number of independent cycles that show the natural variability of production. In the end, both the number of records and the number of markings must be stated separately. The order in which measurements are made can also affect interpretation; an observer who sees the earlier mark may unintentionally lean toward the same decision. The comparison design is prepared so as to reduce this effect. A time display in the image also does not provide an independent reference by itself. If the display is derived from the same unvalidated time base, agreement of the two values may only reflect the shared source. So the link of the reference to the measured event and its adequacy are explained separately. Once a suitable recording arrangement is set up, do not interpret unnecessary enlargement or slow motion as if it revealed detail that is not there. Review may become easier, but visual information that is not in the source does not appear. If the purpose of the measurement requires telling smaller differences apart, the recording rate, view and reference are chosen again. Instead of loosening the acceptance condition after the data are collected, the measurement design must be fitted to the decision required. This approach keeps the usefulness of the digital tool while preventing the evidential strength of the number it presents from being overstated.

Formula