How the RFID smart warehouse management system landed: from live reading to inventory accounting
The following links are to be used for RFID storage modifications: live reading, incident handling, operational verification, inventory services and unusual review. The label is read only to indicate that the equipment has been collected with an identification, which does not allow for confirmation of receipt, determination of exact storage position or deduction of stock. This paper is based on a common high frequency (UHF) passive RFID warehousing application as the scope for discussion, describing how this link is designed, and the conditions under which it needs to be validated at each step.
First define the identity of the item, the status of documentation and changes in inventory
The program design should start with the target audience: whether it is managed in single pieces, boxes or trays, whether to distinguish batches, owner and quality check state, how the connection changes after a box or toddle is removed. Using RFID for bulk collection does not mean that all nodes are equipped with readers; low-frequency operations or non-radio-frequency links can be maintained by bar codes and manual confirmation.
For example, the process can be broken down to a arrival appointment, an inventory of unloaded goods, a review of discrepancies, a physical examination, and an update of the shelf and stock availability. The tag collection on the discharge route can assist in checking to the manifest, but read that the label does not represent a satisfactory quality check; the system needs to distinguish between " already arrived " , " stand up " status, etc., and prevent the non-completed business confirmed material from entering the inventory available at an early date.
The exit and transfer libraries also need to define the conditions for completion. The exit library usually relates to a pick-up mission, which should be sent in and released for confirmation; the migration bank needs to obtain credible target location information, such as a scanned repository number or a validated location recognition device. Ordinary RFID reading events do not in themselves amount to precise positioning, and cannot automatically modify the location of the library simply by "reading it from a reader".
Sensory layer dots: Design around real materials and effective reading windows
The sensory layer includes labels, antennas, readers, triggers and edge treatment software. The design should pre-describe material materials, stacking, labelling positions, channel distances, speed of movement and adjacent operations. Increased launch power may increase side tag reading, but simply expanding coverage does not necessarily improve business accuracy.
When choosing equipment, check local frequency requirements, the tag air interface protocol, antenna ports, available interfaces, and suitability for the environment. Passive UHF commonly uses ISO/IEC 18000-63 or EPC Gen2; check the reader and tag documentation for actual compatibility. A reader's support for a particular tag protocol does not mean it supports the WMS business interface. An SDK, gateway, or middleware is usually needed to connect the two.
The route antenna should be based on the reading of the boundary, labels are placed in the direction and cover, and the collecting window is limited by photovoltaic, geometry or mission command. The sequence of the two triggers can aid the determination, but it still needs to be handled in parallel and reverse situations. Forklifts or shelf schemes need to be independently verified: the peri-point antenna does not guarantee coverage of the whole shelf, nor is vehicle vibration appropriate as the only basis for confirming that removal has been completed.
The label shall be tested with the actual carrier, including metal containers, water packaging, entanglements and the most adverse direction. If necessary, adjust the envelope, fit the gap or place of installation. If only a tray label is read and depends on the back-office box to link it, then make clear that this is an indirect extrapolation of the box list; untamination, reboxing or association failure, cannot continue assuming that the tray label represents all the original goods.
Collapse continuous reading into an interpretable business event
The same label may be read repeatedly in the collecting window by multiple antennas, and the number of consecutive readings does not equal the quantity of goods. The processing of the edge requires that original records are kept, and that the re-reset is generated by the task and the reading window, and that the direction, equipment, time range and abnormal state are recorded for submission to the business system for judgement.
The link between events and operations can be chosen or grouped from three mechanisms:
- Trigger window connection: A collection session is opened by PV or other triggers to limit the candidate label to mission, channel, direction and starting time. The trigger cannot be directly equivalent to a library completion.
- Mission recognition link: The operator selects the receipt or shipment form, collects it and then looks at the list of system discrepancies, confirms the amount of clean-up, removal or business before submitting to this operation.
- Rules verify correlation: compare collections with receivables, due lists, and check label binding, stocking status, operational direction and mission authority; automatically perform inventory services only after the proven conditions have been met.
Identity management should distinguish between EPC code, label entity and business object. The EPC can be coded by rules of practice, not to be considered as a non-removable identity document without authentication; the TID can be used by tag ability when needing an auxiliary verification, but also verify its uniqueness and conditions. Recoverable labels must be untied, re-beltled and life cycle records, history kept in the system retroactively, without having to rely on time stamping to the label.
Abnormal treatment: distinguishing unread, misread and duplicate
The following three types of anomalies can be included in the pilot script, defining the conditions for interruption, compensatory operations, liability for processing and restoration of certification methods:
- Missing: The system lists missing entries, allows for read-and-reading, adjusts the placement or using bar code to clean; the result of the patch is linked to the original task, retains the operator and reason. If the pool is not complete, then check the order and boxing link first, and cannot attribute all differences to radio frequency performance.
- Thread or Non-Target Read: The next-side cargo enters the current collection. Through a trigger window, in which direction the route tasks and readable lists are screened together. RSSI (receiving signal strength) is reflected, shielded and directed, as supporting information, but not a reliable distance scale, and a single threshold does not guarantee that adjacent goods will be excluded.
- Repeated reports or duplicates: Repeated reading, message retrying or repeat clicking may trigger the same operation. The collecting end weighs the session, and the business side also needs to verify the only identifier of the event, document line and status; only control and services such as a thong in the database are submitted successfully to confirm that this inventory change is complete.
The abnormal recording proposal contains event ID, task number, should read, group, difference items, device configuration version and processing trajectory. Key library differences should be confirmed to avoid error release. The reviewers can read, patch, cancel or release as per permission, but each operation should leave the basis; original reading records and final business results should be traced back.
Interfacing with WMS: Disaggregate equipment incidents and inventory services
The interface design identifies which system maintains the inventory master account, then divides the equipment access, and the responsibility for event transmission and business recognition. Usually, border service organizes reading and writing data, WMS implements inventory rules, ERP synchronizes business documents as agreed; actual borders are written into interface files to avoid multiple systems changing the same inventory independently.
- Device access: Linking a reader to the device according to the actual provider ' s SDK, LLRP or other protocol and collecting the equipment ' s online status. LLRP, an lower-level reader protocol, cannot be assumed to support all models; HTTP, MQTT, etc., also needs to be provided directly by the reader or converted from the gateway.
- Event Transfer: Message can contain eventId, challengeId, channel, event start and end time, direction, EPC assembly and configuration version, with antenna and RSSI data if necessary. eventId is used to mark the same event, remain unchanged on retry, and define field units, time format, version, confirmation of response and retention period.
- Business confirmation: WMS verifies whether the documents are valid, whether the material matches the batch, whether the stock status allows for change and is submitted to controlled services. The interface should distinguish between "message received" "check failed" and "business completed" ; only a successful response is received, not because the inventory has been updated.
For example, when the release event is submitted but returns time out, a collection service cannot simply replace an itemId to seize another inventory. The original identification query or retry should be used and the end of the operation should return the result of the process. The offline event is replaced by a clean-up: if the task has been cancelled or the status changed, the current desk account should be rechecked, and not simply overridden by the device ' s time.
Four Boundaries to Verify Before Going On
Accept and accept the reading rate separately from business accuracy.Defines the collection window, the number of target labels and the work situation, then calculates the read rate, the ratio of non-target tags mixed and the time spent processing; separate documents and inventory results are checked. It cannot be generally determined that 98% or 99% is suitable for all warehouses. Even if a live leak is allowed, there must be an enforceable compensation process to prevent the wrong stock from being treated as normal.
Costed by label for a full life cycle.The total cost of ownership (TCO) includes, in addition to equipment and software, labelling, coding, untie, recycling, wear and tear, back-up equipment, on-site construction, system interfaces and continuous transport. One-time labels differ from the cost model for revolving carriers; budgets combine turnover times with manual actions and replacement conditions, and cannot compare only the unit price of the label purchase.
• Involve front-line personnel in the abnormal process test.The test not only allows the operator who knows the system to operate, but also covers normal users in different shifts. See if the difference is clear, whether gloves are useful to confirm, whether the patching adds up or not, and whether unprocessed tasks after a shift. The training should be focused on job assignments and unusual handling, and keep operational guidelines, which cannot be prepared online with a demonstration.
Check the break-net, restart and restore state.Check the power surplus, ground, wire and network coverage, and configure PoE or stand-alone power supply as required by the equipment. The grid is broken to verify local buffers, restart unrecognized events, alert disk capacity and restore back-up transmissions, confirm that no account will be lost or duplicated.
Progress by stage, rather than a commitment to harmonization of work schedules
Project life depends on the size of the warehouse, material complexity, number of interfaces and time available to test on site. It is proposed that five phases be used to organize tasks, each with verifiable exit conditions, re-source resources and shut-down windows to schedule progress.
- Demand and baseline: confirmation of scope of operations, identification of items, inventory status and existing processes, recording of time-consuming and error types of operations, resulting in a mutually agreed list of requirements.
- On-site validation: test tags, points and read borders with real material, covering adjacent corridors, most unfavourable stacking and round-trip movements, original output records and programme restriction statements.
- Interfaces and services interface: from collection to WMS login and upstream synchronization, focus on testing duplicate messages, timeout, disorder, mission cancellation and offline restoration.
- Pilot and discrepancy reconciliation: Select a controlled and representative area, with the results reviewed manually.
- Sub-district switching and handover: as a result of the pilot, only access points, exit conditions and responsible persons are identified, and deliveries are made for backup, interface files, training records and traffic monitoring.
Whether or not to move into the next phase should be based on whether the risk is verified, and the problem of blocking is closed, rather than just the date of the plan. The switch is preceded by an inventory baseline reconciliation and re-engineering exercise, which identifies the differences that can be compensated manually and the circumstances in which the automatic process must be suspended. In the absence of these conditions, the delay extension is usually more manageable than the problem is being carried over line.
The ultimately deliverable RFID smart storage management system can trace an inventory change to business documents, capture sessions, labels, validate rules and confirm results. Along this chain review program, it is possible to judge whether the system really reduces duplicate entry and discrepancy reviews rather than simply shifting manual problems into hard-to-explain automated data.

Ms. Lei
Mr. Pang