How to build an asset management system based on RFID? Architecture design
The RFID-based asset management system is built around four layers of "sensitization-transmit-mediate" architecture, so that the location, state, change into business systems, not rely on manual inventories. This structure applies to multiple scenarios such as manufacturing, warehousing, hospitals, government, and so on, and it's important that pre-scatter needs be solid, and the selection is a match. Here is an application to the implementation sequence.
Take it down first. Don't rush to pick up the equipment.
And so many of the problems -- thinking that RFID was a label and a reader -- came up to the scene, and it was complicated, dense, metal-intensive, and the original scheme was pushed back. So the first step was to sit down and go over the list.
Three things to look at: whether the asset is in dispersed or stacked, whether it's a metal or plastic shell, whether it's static or frequently moving (e.g. tools, instruments, swing gear). Metal surfaces shield UHF signals, liquid environment decays, and these two types of assets are either metal resistant labels or high frequency (13.56 MHz) programmes, which cannot be replicated with generic templates.
And you need to be clear about the functional depth: do only a lot of inventory, or do you want real time positioning? The project that does only an inventory has fewer readers and simple deployments; projects that are going to be located in real time require a large number of reading and writing machines or positioning antennas, which will double their cost and complexity. We see clients start with "real time tracking everything," and then we realize that the budget is far ahead of expectations, and finally adjust to "real time + other regions' time-sets" -- the sooner this adjustment is cheaper.
What does the four-layer structure do?
The whole system is cut into four layers by the data life cycle, each with its own hardware or software component, corresponding to a clear protocol and interface.
Sensory layer: labels, readers, antennas
This is the source of data generation. The labels are attached to assets, the number of the stored asset (usually the EPC electronic product code, similar to the asset's ID number); the reader activates the tag through an antenna that sends a radio frequency signal and reads the data. The mainstream uses ISO 18000-63 (also known as EPC Cass1 Gen2) protocols, which work in 860-960 MHz bands. Passive labels read at distances of 1-10 metres, with active tags ranging from tens to hundreds of metres.
Transfer layer: cable, wireless, industrial bus
Readers pass the read data through web lines, Wi-Fi, 4G or industrial bus to servers. Fixed type readers typically support Ethernet (RJ45) and RS485, partially supporting PEE power supply (a web line simultaneously transmitting data and electricity), making it more economical to set up a plan for IP in the field, avoiding multiple equipment IP conflicts.
Middle Layer: Data Filter and Event Encapsulation
This is a layer that many teams will ignore, but it's really critical. Readers can read hundreds of times a second of the same records, and intermediates are responsible for weighting, filtering, encapsulating standard events like "Affect into Area" and "Affect to Leave" in accordance with business rules, and then sending them to an application system. The EPCglobal defines ALE (Application Level Services) as the main middle of industry.
Application level: business functions and data presentation
The application layer is a direct-to-user segment, which includes the functions of an asset repository, access, return, inventory, location query, depreciation alarm, report statistics, etc.; and also to preset interfaces with ERP, finance, OA systems, allowing the data collected to be automatically synchronized, avoiding "data in RFID system set, "and double-track problems in ERP system set."
Practical experience with labels and type of reader
The selection is not "a rare move" but there are a few details that are often ignored.
The metal asset must be labeled as a metal resistant, and the general label will drop from 5 to less than 0.5 metres. There is a suction material (usually tumbled or magnetic) within the metal tag at more than twice the price of the normal label. A small batch test is required before bulk purchases -- there may be differences in reading and writing performance between different batches of the same type, especially with regard to the import of brands.
Readers: The number of channels, or antenna ports, is a key specification. Fixed readers commonly have four or eight channels. A four-channel reader can connect four antennas to cover a doorway or zone, while eight channels suit dense deployments in large spaces. Verify whether the reader supports the Chinese standard (GB/T 29768-2013) or the international standard, as the two are not fully compatible.
Antenna: A round polarized antenna is suitable for labels that are directed to an unknown scenario (such as a line of cable, which is stacked at will in a warehouse), and a linear polarized antenna is more distant but requires a relatively fixed label. Gains (dbi) and beam angles determine coverage - 8dBi circle polarized antennas are appropriate for near-range wide area coverage, 12dBi orientation antennas are used for remote fixed reading.
The middle is the key to data quality.
The original RFID data has three characteristics: repeat, shake, lose. The same label is read repeatedly as a repetition; the signal fluctuations when the label moves or is masked are shaking; and metal or liquid cover, where human tearing leads to loss. The middle one does three things.
First, deduplicate. Multiple reads of the same EPC over a short period count only once. The deduplication window depends on the setting: it could be one second on a conveyor, or 10 seconds to several minutes in a warehouse.
Second, turn raw reads into business events. For example, “Reader 1 detected Tag A at 10:23:15” is a raw read. Apply business rules to decide whether it means “entry” or “exit” based on the reader's location, or “stop” or “pass through” based on the time interval.
The third is retrospect: data during offline or network shaking are subject to a batch recovery, which cannot be simply discarded. This Shanghai-based Chee-Technology Company stepped through the pit when it was doing plant asset projects -- without a back-up mechanism, the asset desk account would not match and could not be resolved by adding local cache + breakpoints.
The application level is not a big screen, but a business process.
Many of the demands are "three-dimensional visual screens", but what is really a concern for business people is whether the process is smooth, not how the board looks.
Core functions are five: Asset Desks (addition or deletion) , Inventory (handheld or fixed-volume reading comparison), Access to return (with access to the library), Early Warning (overtime, uncounted over time), Report (check accuracy, asset utilization). These five functions run through, and the system is of practical value.
Interface interface is another focus. Asset changes must be synchronized to ERP asset master data in real time or on a regular basis, the depreciation module of finance, and the IT CMDB (Asset Management Database). The interface is usually synchronized with RSTful API or database view, where field mapping is clear at the implementation stage, otherwise later changing field names will be problematic for field types.
A couple of easy-to-touch pits.
First, the site survey may not be thorough enough. After installing antennas, you discover that a metal access gate weakens the signal or that electromagnetic interference from a nearby distribution panel lowers the reader's read rate. A site survey can catch these problems, which are costly to fix later. Bring test equipment and take measurements at every proposed installation point from at least five angles.
The second is to underestimate the implementation cycle. The RFID project generally involves pilot validation (2-4 weeks), batch preparation (see size), interface alignment (2-6 weeks) three phases, and the overall cycle is longer than a pure software project and cannot be managed as expected by the software project.
The third is neglect training. Assets involve a large number of people – warehouse managers, workshop workers, IT traffic, and each role operating scene differently. Without training, the system has been manually modified for three months on the grounds that "sweeping codes are not as fast as hand-filled."
The fourth mistake is leaving no room to expand. Allow 30%–50% spare capacity in reader count, channels, bandwidth, and server configuration so the system can handle business growth.
Summarizing: The RFID asset management system is essentially a four-layer model that matches the asset form, with the intermediates to ensure data quality, and the application level to keep business processes in place. And then, if you do this, the system can be upgraded from "labelled" to "pipe asset."

Ms. Lei
Mr. Pang