How to Choose the Right Instrument for the Question

Why this matters

Shops buy instruments by reading down a specification sheet and picking the better numbers. That works until the day the expensive instrument cannot answer the question and the cheap one could have. Almost every wasted instrument purchase in this trade comes from the same confusion: accuracy, resolution and repeatability are three separate properties, they cost different amounts, and the question you are trying to answer needs one of them far more than the other two. The tech who does not separate them buys digits, trusts the last one, and still cannot tell whether anything changed since spring.

Three properties people call accuracy

• Accuracy is closeness to the true value. It is what you need when the number gets compared against something outside your shop: a nameplate, a code limit, a published threshold. It is the expensive property, because it has to be established against a reference and maintained.
• Resolution is the smallest change the instrument can display. It is what you need when the difference you are hunting is small, and it is cheap, which is why it gets oversold. Adding a digit costs almost nothing and adds no accuracy at all. An instrument whose last digit is finer than its own accuracy specification is showing you a number it cannot support.
• Repeatability is the spread you get taking the same reading on the same unchanged thing the same way. It is what you need when the question is whether something changed, which is most of what a service shop asks. It is the property nobody specifies and nobody measures unless they run a duplicate on purpose (see References).

The consequence worth memorizing: an instrument with a fixed offset and tight repeatability answers "did it change" perfectly and answers "does it meet spec" not at all, and a coarse but accurate instrument does the reverse. Neither is the better instrument. They answer different questions.

Before you take a reading on running equipment

A condition-monitoring question puts a tech's hand on a running machine every visit, so the hazard is in the routine.

• Do not reach past a guard on running equipment for any reading. If the point is only accessible with a guard off, the point is wrong, not the guard. Establish a permanently accessible pad or port while the machine is down and read from it thereafter.
• No loose gloves, sleeves, jewellery or lanyards near a rotating shaft, coupling or belt. A glove that catches pulls the hand in, so a glove is a hazard here rather than a control; the controls are distance, a fixed pad, and a probe with a handle.
• If a reading genuinely requires the guard off or a hand inside, the machine is de-energized, isolated, its stored energy released and the isolation locked under 29 CFR 1910.147 first. That standard covers the mechanical hazard and excludes electrical utilization hazards at 1910.147(a)(1)(ii)(C), which fall under 29 CFR 1910.333(b)(2). A stopped machine also gives a completely different measurement, which is why the accessible-point decision belongs at commissioning.
• Set headphone volume to minimum before applying an acoustic or ultrasonic instrument, then bring it up. These shift inaudible content into the audible band, and a leak or an arc arrives at full gain with no warning.
• Any electrical reading stays behind the energized-work gate at 29 CFR 1910.333(a)(1), with the electrical protective equipment 29 CFR 1910.335(a) requires and an instrument rated for the circuit under 29 CFR 1910.334(c)(3).

Work backwards: five steps from the decision to the instrument

1. Write the decision and the action each outcome triggers. "Above the line we schedule a replacement, below it we do nothing." Skip it and you are specifying capability against a topic rather than a decision, which has no upper bound and no way to say enough.
2. Name which of the three properties governs. Compared against something external, accuracy. Compared against your own earlier reading, repeatability. Compared against a sibling unit right now, repeatability plus enough resolution to see the gap. Skip it and you pay for accuracy on a question that only needed consistency, which is the most common wasted purchase in this trade.
3. Set the required resolution from the smallest difference that would change the action, not from what is available. If a 20 percent change triggers the repair, resolution finer than a few percent of the reading buys nothing. Skip it and you read the trailing digit as information and start chasing movement inside your own noise.
4. Only now look at instruments, and check location and access before the specification. The measurement category rating has to cover where you actually work, and you cannot downgrade the location to suit the instrument. The probe has to reach the point without a guard coming off. Skip it and the instrument arrives correct on paper and unusable at the machine.
5. Check what the instrument demands of the user, and reject anything your shop will not do. Setup time, a laptop, a configuration session, a second person, breaking into a system to connect. Skip it and you own the right instrument and still have no data, which looks exactly like having bought nothing.

Match the property to the question

The fourth row is the one worth staring at. Proving a conductor dead is not a precision measurement, it is a procedure: the live-dead-live sequence in NFPA 70E-2021, 120.5, in the edition your employer's electrical safety program adopts, using a tester suited to the task. A better meter does not improve it.

What deliberately does not go in the bag

• An instrument whose display resolution exceeds its accuracy specification. The trailing digit is not information and will be read as information, particularly by whoever on the crew is newest.
• A second instrument of the same type, bought so two readings can confirm each other. Two of the same design share the same systematic error and the same technique error, so agreement between them is not evidence. Confirmation comes from a different measurement route.
• A precision instrument for a question with a wide band. If the action threshold is a 20 percent change, the tightest instrument on the market and a middling one give the same answers all year.
• An instrument that requires breaking into a system when a non-invasive one lands inside the band. Every invasive connection creates a leak path, a contamination path and a disturbance to what you are measuring.
• An instrument whose measurement category rating does not cover the locations you work in. The one safety exclusion on this list, and not negotiable by technique.
• Anything bought to answer a question measurement cannot answer. Whether a part is the correct part, whether a system matches the design, whether a warranty applies: those are documentation questions.

The buy: two questions about the same bearing

A shop wants to catch bearing wear on its fleet of belt-driven equipment before it strands a customer. Two hand-held vibration meters reading overall velocity are on the table, both illustrative.

• Instrument A: accuracy specified at plus or minus 3 percent OF READING, hand-held probe pressed against the housing wherever the tech can reach.
• Instrument B: accuracy specified at plus or minus 10 percent OF READING, with a mounting pad fixed at each measurement point and a magnetic base that seats on it.

Both bases are stated because both matter: a percent-of-full-scale figure of the same size would not shrink as the reading shrinks, and none of the cancellation argument below would hold on it.

Question one: has this machine changed since last quarter? The action threshold the shop set is a 25 percent quarter-over-quarter rise, so repeatability governs. The shop got the real numbers by running a duplicate rather than reading a datasheet: instrument A came back with a reproducibility of 18 percent, driven almost entirely by where and how hard the probe was pressed, and instrument B came back with 6 percent, because the pad removes the placement variable.

Against A, a 25 percent change sits at 25 divided by 18, or 1.4 times the noise band. Against B, 25 divided by 6, or 4.2 times. The less accurate instrument is the correct buy and it is not close: its fixed 10 percent offset is present in every reading and cancels out of every comparison, which is exactly what a change question needs. Whether one reading is fit for one decision is a separate calculation a sibling article owns.

Question two: is this machine above the equipment manufacturer's published alarm value? The property flips to accuracy, and something more important flips with it. A published alarm value is defined on a basis: a measurement location, a frequency band, an averaging window, a units convention. Instrument B does not report on that basis, so its 6 percent repeatability is irrelevant and no accuracy figure would rescue it. It produces a different quantity that happens to share a unit. Matching the basis comes before matching the accuracy whenever the comparison is against a number somebody else published.

The instrument they did not buy. A multi-channel analyzer with spectral capability would answer both questions and several the shop had not thought of. It was rejected at step 5, not on capability: it needs a laptop, a setup session per machine and a trained operator, and the honest assessment was that it would be used on three machines and then live in a case. An instrument that answers the question in principle and never gets carried has a capability of zero.

What would change the answer. If the shop's work moved toward acceptance testing against published limits rather than trending its own fleet, question two becomes the main question and the pad-mounted instrument comes off the list entirely, however well it trends.

The failure mode. The common version runs the other way: the shop buys A because 3 percent beats 10 percent, runs a season of readings, and finds every machine moving 15 to 20 percent between visits with no pattern. The instrument is fine and its accuracy is exactly as advertised. The shop bought the property it was not going to use, skipped the one that decided the question, and will conclude that vibration monitoring does not work for them.

How to verify you got this right

• Say the question out loud in the form "compared to what." Compared to a published limit is accuracy. Compared to last time is repeatability. Compared to the unit beside it is repeatability and resolution. If you cannot finish that sentence, you are not ready to choose.
• Check the display against the specification. If the last digit is finer than the stated accuracy at that reading, agree as a shop to record one digit fewer.
• Get the repeatability number from your own duplicate run, not from the datasheet, before committing to a threshold that depends on it.
• Walk the measurement point physically before the purchase. Reach, guard clearance, mounting surface, standoff. A specification cannot tell you whether a hand fits.

References

• 29 CFR 1910.147 - control of hazardous energy for mechanical and stored-energy isolation before a guard is removed or a hand goes inside
• 29 CFR 1910.333(a)(1), 1910.333(b)(2) and 1910.335(a) - the energized-work gate, electrical lockout, and the electrical protective equipment required for the exposure
• 29 CFR 1910.334(c)(3) - the requirement that instruments be rated for the circuits and equipment they are connected to
• NFPA 70E-2021, 120.5 - the live-dead-live proving sequence, binding through the employer's electrical safety program or an adopting jurisdiction
• See related: How to Take a Reading Another Tech Can Reproduce; How to Decide Whether an Instrument Is Still Fit for the Decision; What a Meter Category Rating Is Protecting You From