Accuracy, Repeatability and Legal Tolerance: Read the Number Before You Buy a Weigh Filler

Three different numbers get sold as accuracy

Here is the vocabulary I use on a specification, and the question that goes with each term.

• Repeatability is the spread, usually one standard deviation (1σ) over consecutive weighments. Always ask whether the figure is 1σ, 2σ or 3σ: "±0.5%" quoted at 3σ is three times worse than at 1σ, and both read the same on a brochure.
• Mean error is the distance between the batch average and target. It costs money and it is the cheapest to fix, since any filler with a trim function can walk the mean back to nominal.
• Coefficient of variation (CV = σ / mean) lets you compare across fill weights. A CV of 0.5% at 1 kg means σ = 5 g; the same CV at 100 g means σ = 0.5 g.
• Maximum permissible error (MPE) is the metrology term for the instrument itself. Tolerable negative error (TNE) is the per-pack limit on the short side: at a nominal 1 kg the TNE used in the European average-weight framework is 15 g, and no pack may fall below nominal minus twice the TNE.

The consequence for your target weight is direct. Under a minimum-weight regime, where every pack must be at or above nominal, your target sits at nominal plus 3σ. With σ = 6 g on a 1 kg fill that is 18 g per bag handed over free. Under a documented average-weight system with statistical control and feedback you can target nominal plus 1-2 g, because the three packers' rules (average at or above nominal, a sufficiently small proportion beyond TNE, none beyond twice TNE) are evaluated on the batch, not the individual pack. That difference is usually the whole business case.

Five weigh-filler architectures, and where each stops paying

Every powder weighing and filling machine on the market is one of five mechanisms. The table gives ranges I see holding up in production, not brochure ranges.

The most common selection mistake is buying a multi-head for fine powder because the brochure speed is attractive. Fine powder does not discharge cleanly from a radial feeder; it bridges, sticks to the dimple plates, and fills the hopper with dust that settles after the weighment. You end up derated with weekly cleaning. A twin-spout net weigher with a screw bulk feed plus a dribble stage is the honest answer at 500 g to 5 kg.

What wrecks a weigh signal on a real floor

I have never yet found a weigh filler that missed its FAT tolerance because the load cell was bad. It misses because of the plant around it.

Vibration from neighbouring machines

A case packer, a compressor, or forklift traffic on the same slab couples into the weigh frame. Fit isolation mounts on a separate plinth, sized so the mounted natural frequency sits well below the excitation. Never bolt a checkweigher to the same frame as a sealer: I measured a 4 g peak-to-peak swing on a 1 kg checkweigher from a sealing jaw alone.

Air drafts

An HVAC diffuser above the weigh hopper moves a 1 kg reading by several grams, and a dust collector nozzle a metre away does the same. Fit draft shields, redirect the diffuser, and keep 1.5 m between the weigh zone and any open extraction point. Field test: put a 1 kg test weight on the platform and watch it for two minutes with the plant running.

Build-up and mechanical drag

Powder cakes on the discharge flap, the load cell bracket and the hopper walls, and a film of product can hold the flap a fraction open. Specify a silicone or PTFE-coated flap, purge air at the hinge, connections that transmit no force, and an automatic tare check every N cycles so the machine catches its own build-up.

Filter settling time

The digital filter trades noise against speed. A fixed 120 ms window looks fast on a demo and then reads 3 g low because the cell has not settled. Use slope-based settle detection, typically 180-250 ms on a 1 kg net weigher. The mechanical feed is slower than that anyway, so you lose nothing.

Temperature drift

Load cells carry a temperature coefficient on zero and span, and in a 44 °C plant with a 15 K day-night swing that is real. Use temperature-compensated cells, shield them from radiant heat, and run a documented auto-zero at intervals rather than continuous zero-tracking, which hides drift.

Procurement pitfall: a tolerance quoted "at FAT conditions" means conditioned powder and no drafts. Write the SAT tolerance into the contract against your powder and your ambient, with a stated sample size and sigma basis.

Calibration, traceability and the record trail

A weighing machine is only as credible as its calibration history. What I expect in a plant that will pass an audit:

• Certified test weights. A working set to OIML Class M1 or better, a reference set at Class F1 or F2 in the QA lab, certificates traceable to a national metrology institute. Ask for the certificate, not the weight's spec sheet.
• Defined check frequency. Zero check each shift, span check weekly at roughly 50% and 100% of range, full calibration every six to twelve months, and re-calibration after any load cell replacement or mechanical shock.
• Records. SPC charts, reject logs and calibration records retained for a defined period, commonly shelf life plus one year in food.
• Electronic records where applicable. If the HMI writes batch records, a US-regulated customer will expect 21 CFR Part 11 behaviour: an audit trail of who changed the target weight and when, unique user IDs, no shared accounts. Confirm this at enquiry stage; retrofitting access control into an HMI is painful.
• ISO 9001 calibration control. Equipment identified, calibrated against traceable standards, status labelled, with a documented recall procedure when a device is out of tolerance.

Regional notes: EU, United States, Gulf and Southeast Asia

In the European Union, prepackages are controlled under the average-weight and e-mark framework, and the instrument itself falls under the Measuring Instruments Directive (2014/32/EU), with OIML R 87 and R 61 as the international references. A German or Dutch customer will ask for the notified-body certificate and your SPC records before they ask about speed.

In the United States, the device side is governed by the specifications and tolerances in NIST Handbook 44 as adopted by NCWM and enforced by state weights-and-measures programmes, while net quantity labelling sits with FDA and, for meat and poultry, USDA.

In the Gulf, products register with national authorities such as Saudi Arabia's SFDA or the UAE's MOHAP, and legal metrology runs through the national standards bodies, increasingly aligned with OIML. Power is the first practical question: parts of Saudi Arabia run 380 V at 60 Hz while the UAE is 415 V at 50 Hz, so motors, drives and load cell excitation must be specified accordingly. In Southeast Asia, pharma customers work to WHO-GMP and PIC/S expectations and humidity is the enemy: powder that flows in Rotterdam may bridge in Surabaya.

Field note: 1 kg detergent in the Gulf at 44 °C

A detergent packer on the Arabian Gulf side ran a twin-spout linear net weigher at 1 kg, 22 bags per minute, two shifts, six days. Ambient at the filler was 42-44 °C in summer and the supply 380 V / 60 Hz. The plant was convinced the weigher was worn out: mean fill 1,011.8 g with a standard deviation of 6.4 g, so roughly 1.2% of every bag was being handed over.

We found four causes in a day and a half: an HVAC diffuser discharging onto the weigh hoppers, the dust collector's pickup about a metre from the weigh zone, detergent caked on the discharge flaps holding them a fraction open, and a fixed 120 ms settle window that took the weighment before the cell had settled.

The fixes cost less than a new machine: relocate the diffuser and fit draft shields, move the extraction pickup, fit coated flaps with hinge purge air, slope-based settle detection at 220 ms, temperature-compensated load cells, and a weekly span check against certified M1 weights. Afterwards, mean fill was 1,003.4 g with σ = 1.9 g: 8.4 g recovered per bag, about 44 tonnes a year, payback under eight months. The load cells were never the problem.

Metrology questions we handle on enquiry calls

Is ±1% the same as a CV of 1%?
No. Ask which sigma the figure refers to, over how many weighments, with what product and at what fill weight. A serious supplier sends the raw data set, not just the percentage.

Can I run closer to nominal if I have a checkweigher?
Yes, provided you also have the records. Feedback plus documented SPC is what lets you use the average-weight system instead of padding every pack to nominal plus 3σ.

Do I need a multi-head for powder?
Usually not for fine powder below 1 kg. A multi-head earns its keep on free-flowing granules. For cohesive powder, an auger with load-cell trim or a net weigher with a screw bulk feed is more predictable and easier to clean.

How often should a powder weighing and filling machine be calibrated?
Zero check each shift, span check weekly with certified weights, full calibration every six to twelve months, and immediately after any shock or component change.

What should the FAT and SAT actually test?
FAT on the vendor's floor with conditioned product, at least an hour of running, 30 consecutive weighments per spout, mean and σ reported. SAT on your floor with your powder and utilities, same statistics, over at least two hours. Agree both sets of numbers in the contract before the machine is built.

What about spares and local service?
Load cells, flap actuators and feeder drives are what fail. Ask the lead time for each, whether the HMI and manuals are genuinely in English, and who can be on site within a week. I have seen a plant lose more to a three-week wait for a load cell than the checkweigher ever saved.

Written by Helen Xu | Chief Industrial Application Engineer

Helen Xu is a Chief Industrial Application Engineer with 9 years of specialized experience in packaging machinery and liquid filling machine design, equipment model selection, and full production line process optimization. He focuses on delivering customized packaging & filling solutions for pharmaceutical, food, and chemical manufacturing industries, with mature practical expertise in GMP compliance, ISO 9001 quality management standards, and turnkey large-scale filling & packaging production line integration.