How does the RFID fixed asset management software architecture work?
The RFID fixed asset management system architecture is structured along five layers, with each layer of the system being divided into "sensor-transmit-process-appliance-storage" and each layer of it being functionally functional, evolving independently. The sensory layer is responsible for reading and writing labels, the transfer layer is sent to the middle, the processing layer is filtered and event aggregates, the business layer achieves a table of contents, positioning, depreciation, etc., and data layer ensures that the entire life cycle of asset records can be traced.
Why must the RFID fixed asset system be stratified?
The fixed asset is a million pieces, each of which binds to an RFID electronic label, and the system collects data in millions of degrees per day. If all logic is mixed in one program, single-point failure will cause the whole inventory chain to collapse. When we plan for a large manufacturing firm, we initially plug all our functions into a single-body application, and then the CPU is filled with more than 500 labels read and sent out; Then we break the architecture down into a sense, gather it, middle it up, four levels of business, each one being extended horizontally, and then solves the problem.
The benefits of tiereding are three: one is that layers of technology can be independently selected, such as C++, with a sense layer driven by Java or Python; two is fault-segregated, and the card failure does not affect the operation of the financial depreciation module; and three is lateral upgrades, so a new RFID label will need to change the sensory layer driven without moving the business code.
What's the whole structure?
The standard stratification structure consists of five layers:
- Sensory layer (RFID hardware layer)Electronic labels, fixed-style readers, handheld terminals, RFID printers, antenna arrays.
- Collecting transfer layer: Reader intermediate drive, TCP/HTTP/MQTT protocol access, data aggregation gateway.
- Data-processing layer (middle level): Label weight, event filtering, data cleansing, rule engine, equipment management.
- Operational application levelAssets are: inventory, receipt, transfer, end of life, depreciation, location, statement.
- Data storage layer: Master data, label binding, event flow, operation log.
The five layers are docked through a standardized interface: the sensor layer to the collector layer by the EPCglobal ALE (appliance level event), the collection layer to the processing stage by MQ (message line) message middle, and the RESTful API between the business and data layers by a common network interface style.
How do you choose and deploy the sensory layers?
The key to the sensory layer is a match between labels and readers. First, labels: Metal assets are marked against metal RFID tags, frequency bands are given priority over UFF (UHF, 860-960 MHz), with reading distances of up to 8-12 metres; normal office assets can be labelled using plain paper labels, at a unit price of as low as 0.3-0.5 yuan/min. Readers have two types: Fixed-readers are installed on the entrance to and exit from the machine room or on the shelf, and the count is large; Handheld terminals are used for manual sampling and remote asset search.
Antenna density and angle matter. A standard records room of about 50 m² can use two to four circularly polarized antennas with a 4–6 m coverage radius. For equipment rooms with densely packed cabinets, use one linearly polarized directional antenna per aisle to minimize interference. Survey the electromagnetic environment first: metal shelving, microwaves, and elevator variable-frequency drives can affect read rates.
How do you ensure that the data is stable?
Readers to intermediates are usually connected with TCP long times, because RFID labels are high-frequency operations, and thousands of EPC data can be generated per second. The middle is supported by multiple reading writer and sent into the system, typically using Netty or Mina frames for a different IO (a high-efficiency communication that does not block the main line), and single servers can carry 200-500 readers.
The web layer suggests that the message buffer be made by MQTT or Kafka. MQTT is suitable for a small bandwidth, with a large number of devices, such as branch count data assembled in the local Broker; and Kafka is suited to a big data volume scene, which can be up to a million-square-second level per second. Both are designed to maintain information and avoid loss of data when the network shakes. We use Kafka, topic, in actual projects, to follow up on asset-type ID+ event-type partitions, to trace the entire life cycle.
What do you need to do on the data-processing level?
The middle layer is the brain of the entire structure, which is basically four things:
- Tabs to Heavy: Multiple antennas may read one tag several times in one second. Deduplicate by EPC to avoid generating large numbers of redundant events.
- Organisation: White and blacklists are created, unregistered labels are dropped directly and invalid data pressure is reduced.
- Data Cleaning: Dismissing a signal whose reading rate is below the threshold, such as RSSI (the greater the value indicates the signal ' s strength) data below -65dBm may be considered unreliable.
- Rule engine: Define business rules. For example, alert if a tag leaves its authorized zone for more than five minutes, or flag repeated moves of the same asset within one hour as an exception.
This layer is recommended for construction with Java or Spring Boot, which matches the rules of Drools or Avator and handles delay control within 200 ms.
What core modules are included in the operational application level?
- Asset master data management: Asset number, name, specification, department, date of purchase, original value, net value, method of depreciation.
- Label binding and untie: Print labels, write to the EPC, bind to asset numbers and tie relationships stored in the only index table of the database.
- Smart Point: Supports the automatic generation of inventory tasks by region, by sector and by asset class, with results automatically compared to differences.
- Assets flow(b) Access, loan, return, transfer, assignment, and recording of operator and time throughout the process.
- Depreciation and end-of-life: Interlocking financial systems, depreciation on a monthly basis and freezing label status at the end of the asset.
- Visualized reports: asset distribution hot drive, idle ratio statistics, inventory accuracy statement.
The frontend recommends a Vue or React framework, moving the peer point App to use the Flutter or Native Development, Scanning +RFID double-reading, which will enhance on-site operation efficiency.
How do you design the data storage?
Use three types of storage. A relational database such as MySQL or PostgreSQL holds asset master data and tag associations. A time-series database such as InfluxDB or TimescaleDB stores the real-time event stream from readers. Object storage such as MinIO holds inventory photos and scans. Keep asset activity logs in a separate database, partitioned by month, and retain them for at least five years to meet audit requirements.
Data backups use daily full backups and hourly incremental backups. A primary-replica architecture separates reads and writes. The recovery point objective (RPO), the point to which data can be restored, is within one hour, and the recovery time objective (RTO), the time needed to restore service, is within four hours.
How can system security and reliability be guaranteed?
The RFID system has several specific security risks: the label is torn out of rewriting, the reader is torn off, the wireless signal is to be monitored. The response is to: the tag follows the EPC Class 1 Gen2 standard and access password is enabled, the business data encryption library is to be decoded, the reader and intermediate are to be certified in two directions using TLS (transmit layer safety protocol).
With regard to reliability, the key reader suggests double-heat, with core business modules deployed on at least two servers, and database owners having a switch time of no more than 30 seconds. The inventory peak period allows temporary amplification of the collection nodes, and operational layer non-state service directly adding machines.
What are the most easily stepped-in pits at the time of implementation?
Three common problems: first, the label selection does not take into account the material of the asset, and metal equipment has a normal tag reading rate of less than 50 per cent, which must be replaced with metal tags; secondly, antenna deployment is too close to the metal ceiling or wall, leading to a serious signal reflection, with read distance down from 8 metres to within 2 metres; and thirdly, new assets and end-of-life processes are not tied to the label, resulting in assets in the database without RFID labels, and large area leakages at the time of the count.
The method of circumvention is to perform a small range of POC (conceptual validation) before implementation, and test the read rate with real assets; and to embed tag binding processes into asset repositories so that one thing is bound.
In summary, the RFID fixed asset management system architecture is designed to be sensory, collecting, processing, operating, storing five layers of the layer, each with a standardized interface, and extending in an independent horizontal way. From selection to deployment and then to the location of the business module, each step requires a specific analysis of the physical environment and use of the asset mass. The stratification of Shanghai-based Cheeky Way Technologies in the multi-industry RFID asset inventory project allows the system to reach a level of one million-person assets, while ensuring an accuracy rate of more than 99%. Then if you need to extend to asset positioning or smart warning, just add rules on the middle floor, and modules on the business level, without reconfiguring the bottom structure.

Ms. Lei
Mr. Pang