Why Your Precision Measurements Keep Drifting
I've spent more years than I care to count watching people waste money on expensive measurement equipment and then get results that look right on paper but fail in production. Walter Lloyd Higgins understood this problem better than most of his peers. He wasn't interested in theory for its own sake. His work at the National Physical Laboratory in the mid-twentieth century was driven by one simple observation: the gap between what you think a part measures and what it actually measures is where everything falls apart. The approach Higgins developed for isolating error sources in industrial measurement systems is still relevant today, even though nobody talks about it by name anymore. Most engineers have probably heard of him through textbooks on the history of metrology. Very few actually apply his methods. That's the problem I want to address here.
Walter Lloyd Higgins Pioneering The Science Of Precision In Industrial Application
What Higgins Actually Did Differently
Before Higgins started working on this systematically in the 1940s, precision measurement in industry was basically a collection of hand-me-down practices passed from shop floor to shop floor. You measured something with a comparator, checked it against a gauge block, and crossed your fingers. There was no real framework for understanding which errors came from the instrument, which came from the operator, and which came from the environment. They all just mixed together and you got a number that meant nothing reliably. Higgins' contribution was treating measurement uncertainty as a structured problem rather than an annoyance. He developed a methodical way of separating systematic error from random error and assigning each a measurable influence on the final reading. This seems obvious now. At the time it was considered overcomplicating things. I've seen senior quality managers push back against the same idea in 2018. Nothing changes much in this industry. The core idea is straightforward enough that I can explain it without resorting to page after page of equations. You take a reference standard, which is just a part you trust, and you measure it repeatedly under different conditions. Different temperatures. Different operators. Different instruments if you have access to multiple. Then you look at the spread and bias across those measurements. Where the spread increases predictably, that's your sensitivity to that variable. Where there's a consistent shift, that's a systematic error you can correct for.
How To Apply This on a Real Shop Floor
Here's the practical version. Let's say you're measuring the diameter of a shaft using a pneumatic comparator. The specification calls for a tolerance of plus or minus five microns. Your process capability looks fine on paper. But your shipping rejects are running at about four percent. Something is wrong and you can't find it by looking at individual measurements. The first step is to establish a measurement baseline. Pick a single artifact -- a certified gauge block or a master ring gauge -- something stable and traceable. Measure it twenty times in a row without adjusting the instrument. Record the average and the standard deviation. This gives you the repeatability baseline for that specific instrument in its current condition. Then introduce controlled variation. Change the ambient temperature by five degrees and repeat. Swap operators and repeat with the same artifact. If you have a second comparator, run the same artifact through it. Each combination of conditions gives you a data point. After about eight to twelve combinations, you start seeing patterns that tell you which variables matter.
Get the Full Details
In my experience, temperature is almost always the biggest offender if your environment isn't climate-controlled. A three-degree shift in a steel shaft measurement setup can account for two to three microns of apparent dimension change. Operators tend to introduce bias through how they position the part or how consistently they seat it against the measurement stops. And when you bring in a second instrument, the calibration offset between the two usually shows up clearly as a systematic shift. Once you've mapped out the error sources, you apply corrections. Temperature compensation goes into your measurement software. Operator consistency improves with better fixturing. And instrument calibration offsets get baked into your data logging system so every reading is already adjusted. This typically reduces your measurement uncertainty by half, sometimes more depending on how sloppy your original setup was. Capping the process down from an uncertainty band of maybe ten microns to around five, which finally makes your four percent reject rate make sense instead of being a mystery.
The Counter-Intuitive Part Nobody Teaches
Most people approach measurement improvement by buying better equipment. That's the third-most effective thing you can do. The second-most effective is environmental control. The most effective is understanding the error structure of your existing setup and working within it. Higgins emphasized this repeatedly in his publications. A well-characterized mediocre instrument will outperform a poorly characterized high-end one every time in production environments. Another thing beginners consistently miss: measurement uncertainty doesn't shrink linearly with more data points. Going from ten measurements to twenty might cut your confidence interval by twenty percent. Going from fifty to a hundred will buy you maybe five percent improvement. There's a hard limit to how much averaging helps once you've addressed the dominant error sources. The sensible move is to spend that measurement time on controlling temperature and standardizing fixturing instead of running more repeats for the sake of looking rigorous. I hit this wall personally on a project involving thin-walled aluminum housings. We were measuring bore concentricity with a coordinate measuring machine. The specs were tight enough that we needed sub-micron repeatability. We threw everything at it -- longer measurement cycles, more points, a temperature-controlled room, multiple probes. The results kept drifting in ways that couldn't be explained by any single variable. What finally resolved it was recognizing that the aluminum was still releasing machining stresses during the measurement wait time. The part was literally changing shape while we were trying to measure it. We ended up using a quick-check pneumatic system for acceptance sampling and only used the CMM on sample artifacts held at thermal equilibrium. The CMM became a calibration tool rather than a production measurement tool, and the whole process stabilized.
When This Approach Falls Apart
This framework assumes your error sources are stable and repeatable. That's a big assumption. If your measurement process is affected by something chaotic -- tool wear on a milling machine that's unpredictable, material variation in raw stock, operator rotation that you can't control -- the error structure itself becomes unstable and the corrections you apply become obsolete within weeks. In those situations, the better move is to reduce variability at the source rather than trying to characterize and compensate for it downstream. There's also a practical limitation that catches people off guard. This method requires access to stable reference artifacts and enough time to run the characterization experiments. A small job shop with one fast-moving line and ten people cycling through setups may not have the bandwidth to dedicate a week to measurement system analysis. In those cases, a simplified version focusing only on repeatability and reproducibility -- what the automotive industry calls GR&R studies -- will get you most of the benefit with a fraction of the effort. If your tolerance bands are wider than ten percent of the nominal dimension, none of this matters much. The measurement uncertainty is a rounding error compared to the process variation. You're better off investing in process control than measurement refinement. The industry tends to apply precision measurement techniques uniformly regardless of whether the tolerances justify the effort, which is why so many shops end up with expensive equipment and mediocre results.
Starting Without Overcomplicating It
You don't need a laboratory or a team of metrologists to apply the basic principles. Grab a certified reference artifact. Measure it. Note what changes when you change the conditions. Correct for the systematic stuff. Ignore the noise until it exceeds your tolerance band. That's basically the entire methodology distilled down. The people who get the best results treat their measurement system like a process variable rather than a passive tool. They know its limitations, they account for them, and they don't trust numbers that fall in the gray zone near a tolerance boundary without additional verification. Higgins' work was always grounded in that practical mindset. The theory was secondary to making sure the part was actually what it was supposed to be.
