RFID in practice

Counting 160,000 items – the bachelor thesis on RFID

An annual stock count that takes two to three months with two full-time equivalents. Can RFID cut it? This is my bachelor thesis from 2020 – the warehouse in numbers, what the pilot test actually showed, why tags on metal stopped answering, and what we could not answer.

Back to the episode Arrow keys or swipe

What this covers

The warehouse the thesis was about, in numbers
Why an annual stock count takes two to three months
What the 2019 pilot test actually showed
Why the same tag gets read seven to ten times
Metal, grounding and the chips that stopped answering
Active or passive – the two scenarios
What the thesis could not answer

What this covers (cont.)

What the conclusion said, and what it did not

The warehouse, in numbers

The numbers are half the answer to why this is hard.

The supply base, as it was in 2020

Measure	Figure
Material numbers	12,000 distinct
Articles	up to 160,000
Stock value	up to 500 million kroner
Share held as safety stock	around 80 per cent
Turnover	around 20 per cent
Annual count	2–3 months, two full-time equivalents

The last two rows belong together. When four fifths of the warehouse is critical components kept there so they can be shipped out the day something breaks offshore, you spend every year counting items that have not moved. Some suppliers have stopped making the part, so you buy a couple of spares and put them on the shelf. There they may sit for years.

The paper

What surprised me most when I had to describe the existing situation was how little of the problem was about people.

The errors did not come from carelessness. They came from the distance between the shelf and the keyboard.

What we found: the warehouse had minimal errors – in the count, at goods receipt, on returns from offshore, on location changes. The staff we interviewed said corrections were rare, and that when one was needed they traced the cause rather than just fixing the number. The work was not done badly. But when a warehouse worker has several tasks to finish outside before returning to the office, the scope for human error grows – without anyone having done anything wrong.

The pilot test

Here we were lucky. In the summer of 2019, half a year before we started, the company had already run a pilot test on passive UHF tags. We were given access to it, and could build on its findings instead of guessing.

The pilot setup

The tagsPassive UHF tags attached to a wide variety of goods in the racking.
The read head on the palletFixed in place, raised and lowered through the rack by forklift.
A counting system of its ownHad to be built from scratch. It registered tags against material number and location, and told the read head which location to count.
FilteringReduced the many reads of the same tag to one item, and showed "at shelf" against "in stock".

The third step is the point. The tags and the reader could be bought. The counting system did not exist and had to be written – even for a pilot meant to answer a single question. This is what tends to be underestimated in RFID projects.

The one read seven times

This is the detail I remember best, because it was not in any of the textbooks we read first.

A read head does not count items. It registers signals – and it registers the same signal over and over.

What it actually looks like: as long as the read head is within range of a tag, it reads that tag again. In the pilot the same ID was read seven to ten times per count. Every tag has its own ID, so the counting system's job was to strip out the duplicates, so that one tag read ten times was booked as one item. Without that filtering the warehouse would have "grown" eightfold every time somebody counted.

The metal

The pilot had one real problem, and it was physics.

The two scenarios

The choice was between active and passive RFID, and it was not capability that decided it.

Active or passive, for this warehouse

Scenario 1

Active RFID

Battery in the tag, range of several hundred metres
Would in practice give a live stock level at all times
More expensive chips – and the warehouse holds up to 150,000 articles
Every battery has to be replaced after a few years

Scenario 2

Passive UHF

No battery – the tag is woken by the reader's radio waves
Smaller, cheaper, and a long service life
Six to eight metres is enough inside a rack
Applied much the way a barcode label is today

Active RFID fell on the arithmetic, not on performance. A warehouse with 150,000 articles is 150,000 batteries that will one day need changing, and at that point the labour costs more than the chips. The last bullet on the right was an argument in itself: goods are already labelled with barcodes, so the labelling step already exists in the process.

What we could not answer

This is the part I would give the most weight if I were writing the thesis today, because it decides whether the arithmetic holds.

Three open questions

The labelling jobEvery article has to be tagged before you can count with RFID. We could not estimate how long it takes to tag one rack, and therefore not what the one-off cost actually is.
MultipacksA box of 100 parts is booked as 100 units in SAP, but goes onto the shelf with the parts still in the box. Nobody tags every single screw.
Power and networkThe solution depends on both. An outage needs backup power and local storage – otherwise the count stops.

The middle point is the most interesting, because better technology does not make it go away. One option we sketched was to mark the tag with a note that the contents must be counted by hand, or to treat an unopened pack as whole and only count the one that has been opened. Both are working routines, not technology.

What the conclusion said

And just as importantly: what it did not say.

The conclusion was conditional. That it could pay off, and that the company could consider investing – not that the gain had been calculated.

The concrete estimate: the count as it stood, two people for two to three months, could be replaced by one person driving a forklift from rack to rack. The time freed up was the main argument – not because counting is valuable in itself, but because staff can then spend the time catching errors early. During our visit an alert employee stopped an active order that was no longer needed. That is the kind of intervention that pays for itself, and it requires somebody to have the time.

What the thesis taught me

About RFID: that physics decides, and that it decides early. Metal and liquid are not details you solve at the end.

Key points

The warehouse held 12,000 material numbers and up to 160,000 articles, with a stock value of up to 500 million kroner.
The annual count took two to three months with two full-time equivalents – not because the work was poor, but because the method was manual.
Around 80 per cent of the warehouse was safety stock. Much of what was counted had not moved in years.
The pilot read passive UHF tags at six to eight metres through the racking – on cardboard, wood and plastic.

Key points (cont.)

The same tag was read seven to ten times per count. The system had to filter those reads down to one item.
Tags stuck straight onto metal were grounded and stopped answering. That was the pilot's only real problem.
Active RFID was ruled out on cost and battery replacement, not on capability.
Multipacks could not be solved. Nobody tags every single screw.

Key points (cont.)

The conclusion was that it could pay off, with reservations – not that it would.

Questions this answers

How long does an annual stock count take in a large warehouse?
What did the pilot test of passive UHF RFID show?
Why must an RFID counting system filter out duplicate reads?
Why does an ordinary RFID tag stop working on metal?
What is the practical difference between active and passive RFID?
Why was active RFID ruled out?

Questions this answers (cont.)

What is the problem with counting multipacks?
What did the thesis conclude?

Sources

Jan Sindre Heltne and Stian Lunde – «Bruk av RFID-teknologi for effektivisering av varetellingen på lageret» (Using RFID technology to make warehouse stock counting more efficient). Bachelor thesis SCM600, BSc in Logistics and Supply Chain Management, Molde University College, 2 June 2020. Supervisor Bjørn Jæger. In Norwegian.

Spørretimen

Read the whole thing at