The conveyor as a read point: designing for a read that happens while the goods move

01 / FIELD NOTE

Keep the decision tied to the operating context.

A conveyor is a read point with a property no other read point has: the goods do not stop, and nobody is watching the read. At a dock door a missed tag can be found by walking the load. At a packing bench the operator notices that the screen did not update. On a conveyor the carton is already downstream by the time anyone could notice, and the record simply does not contain it. That asymmetry is what makes an in-motion read point a design problem rather than an installation.

The physics of the read are a function of time. A tag entering the antenna field needs a certain amount of energy before it can answer, and a reader needs a certain number of exchanges to be confident of the identity. Both take time, and the time available is set by the belt speed and the length of the field along the direction of travel. Doubling the belt speed halves the time; narrowing the field does the same. Neither of those is a software setting, which is why an in-motion read cannot be fixed at the reader once it is failing.

The field along the belt is therefore the quantity to design, and it is longer than the physical antenna. A directional antenna’s usable field extends some distance in front of it, so the effective window is a length of belt rather than the face of the antenna. Lengthening it — with a second antenna, a different antenna, or a lower belt speed — is what buys read attempts, and it is worth stating the target as a number of attempts per tag rather than as a read rate, because the attempts are what the design controls.

Orientation is the second lever, and on a conveyor it is unusually controllable. Cartons on a belt are typically aligned, so a linear antenna and a consistent label position can be made to hold — which is the arrangement that gives the longest read for the least power. The exception is the mixed load, where parcels arrive in arbitrary orientation, and there a circular antenna trades range for tolerance. The presence of even a minority of oddly-oriented items is what decides between the two.

What is in the box matters as much as what is on it. A tag inside a sealed carton reads through the cardboard, but the contents interact with the signal: liquid absorbs it, metal reflects it, and a dense packing of many tags produces collisions that a single tag would not. The practical consequence is that a read point validated with test parcels can perform differently on live goods, so commissioning against the real product mix is not optional.

A miss has no single meaning, which is the part that has to be designed for rather than measured. A carton that produced no read may have an absent, damaged, shielded or unreadable tag; it may have been read while the field was occupied by a burst of others; or the label may be on a face that never pointed at the antenna. Those causes are indistinguishable from the record alone, and treating them as one category is what turns a read point’s exceptions into background noise that staff learn to ignore.

The recovery path is what makes the difference between a read point and a bottleneck. If the only outcomes are a clean read or a silent gap, the system cannot be trusted with anything that matters. Making the miss visible — rejecting the carton to a rework lane, raising it to an exception queue, or reading it again at the next checkpoint — converts an invisible failure into a handled one, and the choice between those depends on whether the belt can be stopped and whether a person is present.

Duplicate reads are normal and need a rule. A tag in the field reports continuously, so one carton produces many observations. The system has to decide which of them constitute the event: the first, the strongest, or a set collapsed into one once the tag has been absent for a defined interval. Without that rule, the same carton either appears many times or is counted by whichever read happened to arrive first.

The direction of travel should be used, not inferred. Because the goods move predictably along one axis, a read point can tell entry from exit — which is what distinguishes a carton moving forward from one that has jammed under the antenna and is being read for minutes. A jam produces a stream of reads from a stationary tag, and a reader that reports occupancy rather than crossing will report a great deal of activity for no movement at all.

Where the read point sits in the process decides what can be done with a miss. A read at induction happens early enough that an exception can be pulled from the line without disrupting anything downstream. A read before sortation can redirect an unreadable parcel. A read at dispatch is a final check, and a miss there has already missed its chance to be corrected. The value of an in-motion read point comes mostly from how early it is.

There is a limit worth accepting: an in-motion read verifies identity, not completeness. It can confirm that a carton with a known identifier passed, and it cannot confirm the contents of that carton. Where the contents matter, the answer is either to read the individual items at a point where they are separated, or to establish the contents when the carton was packed and treat the conveyor read as a movement event. Both are legitimate, and they produce different records.

The design is therefore settled by three numbers and one rule: the time a tag spends in the field, the number of read attempts that gives, the real product mix the field will see, and what the system does when a carton produces nothing. A read point specified against those is one whose misses are known, countable and recoverable. One specified against a read rate quoted for a bench test is one that will be discovered in production, by the absence of something in a record nobody is watching.