A warehouse management system knows what it has been told. It records a receipt when a pallet is scanned in, a movement when a task is confirmed and a pick when an item leaves a location. When those events are entered by people, they are late, occasionally wrong and sometimes missing, so the record drifts away from the physical stock and needs periodic counts to correct it. Sensor networks aim to close that gap by observing the physical world directly and reporting changes as they happen. The result is not simply a faster version of manual scanning. It changes the design of the operation, because decisions about replenishment, slotting and order release can use a picture of stock that is current instead of last night's.

Which sensors observe the stock?

Several technologies are in common use, each with different strengths. Barcode scanners are cheap and accurate but need a person or a fixed station to present each item. Passive UHF RFID tags can be read in bulk without line of sight, which suits dock doors and portals where whole pallets pass through. Active or Bluetooth Low Energy tags can report location continuously at higher cost. Weight sensors under shelves or bins infer pick events from changes in load, and cameras can recognise items or confirm presence. Most warehouses combine several, using the cheapest technology that gives the needed confidence at each point in the flow, rather than trying to instrument every item with the most capable one.

Advertisement

Moving and standardising the data

Raw reads must travel from readers and gateways to the systems that use them. MQTT, a lightweight publish and subscribe protocol, is widely used for this, because devices can publish events to a broker without knowing who will consume them. OPC UA is common for exchanging data with industrial equipment such as conveyors and sorters, and it carries a rich description of what each value means. For identity and event sharing across organisations, GS1 standards define how items are identified and how events such as receiving and shipping are described, for example through the EPCIS specification. Using standards at these boundaries keeps a warehouse from being locked into a single vendor's way of describing its stock.

Noisy reads need reconciling

Raw sensor data is not inventory. RFID readers report the same tag many times, miss tags that are shielded by metal or liquids, and sometimes read tags on neighbouring pallets. Weight sensors drift and need calibration, and cameras misjudge cluttered shelves. A middleware layer therefore deduplicates events, smooths them over time and applies rules, for instance treating a tag as moved only after it is seen by a second reader. The cleaned events are then reconciled with the warehouse management system, which remains the system of record, and discrepancies are flagged for cycle counts rather than silently overwritten. Treating sensor output as evidence to be weighed, instead of as truth, keeps one faulty reader from corrupting the whole record.

Why it matters

Accurate, current inventory affects almost every warehouse metric, including pick accuracy, stockouts, order cycle time and the labour spent on counting and searching. Sensor networks provide the observations that make that accuracy possible, but the benefit comes from the whole pipeline, which includes sensible placement of readers, standard data formats, filtering of noisy reads and disciplined reconciliation. Teams that invest in the pipeline as well as the hardware get an inventory record they can act on.

A sensor tells you what it saw. Inventory is what you decide to believe about it.