How is the architecture of the asset RFID management system designed? Hardware and software synergies
Start with a count.
Last year we helped a machine shop with annual inventory. The tool workshop alone took three days. Four people counted rack by rack, wrote down results and entered them in Excel, and the books still differed by more than 30 items. After deploying RFID, which uses radio waves to identify tags automatically, two people completed the same walk with handhelds and had results in two hours. Discrepancies fell into single digits.
The system is not really complex: the bottom is a RFID label, a reader, an antenna-based sensory layer, which is the middle for data aggregation and filtering, the upper level is the asset table, positioning, process application. Three layers communicate with standard interfaces, one is connected to events, and either level is disconnected from the system into "half-finished." Here's how this structure is to be broken down.
Sensory layer: labels, readers, antennas
The sensory layer is the "nero-end" of the entire system, which is responsible for making assets identifiable.
The label is not as expensive as it is.
Use on-metal tags on metal. A standard tag mounted on a metal enclosure can drop from a 5-6 m read/write range to a few dozen centimeters and may be nearly unreadable. For plastic, wood, and paper assets, ordinary passive UHF tags are sufficient and typically cost from a fraction of a yuan to a little over CNY 1 each. Use liquid-resistant tags around liquids because water absorbs RF energy. Battery-powered active tags are more expensive and last three to five years. They generally suit high-value equipment that needs to report its location, such as large hospital instruments and factory molds.
The label is also sealed. The plant environment is heavily oil-contaminated, at least IP65 (complete dustproof, low pressure water column flushing); the equipment used outdoors is marked with IP68 (dustproof, long-duration inundation). For chips, Impinj Monza, Alien Higgs series, storage capacity varies from 96 bit to 512 bit, and a full marking of 96 bit is sufficient, if you want to place a date check on the label, then select 512 bit.
Readers and antennas set by scene
Fixed-readers are attached to external antennas, usually installed at the warehouse entrance and exit entrance, and on both sides of the tunnel form an "identified door." The radiation distance from the antenna is strongly related to the field environment, with open corridors having a view of 8-10 metres, with a shelf density of typically between 3 and 5 metres, and metal-covered environments likely to fall to 1-2 metres. So, when designing the programme, it is done in situ, without seeing the label parameters on the product manual.
Handheld terminals are suitable for motorized counts, such as assets at the bottom of a shelf, equipment that is temporarily deployed. The prevailing handheld terminal can now read from between 1 to 8 m, with a continuous flight of 8 to 12 hours, and can carry 1.5 m down. There are two modes of synchronization of data from the handheld end and backstage: real time uploads to Wi-Fi or 4G/5G, and offline mode saves local, return to workspace and load, and export formats are usually CSV or Excel.
Intermediate: How hardware data becomes softwareable information
The middle is the "translator" between hardware and software. A reader can report hundreds of original read-card data per second, mostly noise -- the same label is read over multiple antennas, reading around the system's not-release tag, empty reading. The first thing that a midway is to filter and remove weight.
Three Core Rules for Data Filtering
The first is to remove the weight: using the label EPC (electronic product code, equivalent to the asset's only ID number), to keep only one of the strongest signal records in a few hundred milliseconds of time window. The second is to filter the strangest tag: to keep only the EPC registered in the asset desk account, unregistered for direct disposal, and avoid misreporting. The third is to trigger an event: not every time you read it into the library, but when the label enters or leaves a geographical fence (a virtual area that is co-circled by one or more readers), it triggers an asset move event.
The intermediate is usually deployed on the edge of the site or back-office server, providing external information about RESTful API (a HTTP based interface code) or MQTT (a lightweight Internet protocol). The commonly used medium is Impinj ItemSense, the Open-source Apache Kafka Streading, and a team based on Kafka. Select which data and real-time requirements - per day reading cards below million-grades - will be enough with open-source solutions.
Application: What does it matter?
The application layer is a direct interface for asset managers, with four core functions broken down: asset desk accounts, real-time positioning, mobile trajectory, and life cycle management.
Asset desk accounts are the basis
Each asset has a digital record with its name, category, purchase date, original value, current custodian, location, and status (in use, idle, or disposed of). Key fields link to the RFID tag's EPC so a tap or scan retrieves them. When an asset is added, the system generates an EPC and prints a tag. Combined QR and RFID tags, with a printed QR code and embedded chip, are common and work with handhelds and phones.
Real-time positioning and moving trajectory
Positioning accuracy depends on antenna density. Placing a reader every 6-8 m in a warehouse and using an RSSI (received signal strength) algorithm can locate assets to a specific rack. In densely racked areas, a directional antenna at the end of each rack can narrow the error to 1-2 m. GPS can support positioning over large outdoor areas, but it works only for outdoor equipment and uses more power.
The move trajectory is a time-bound routing of each location event. This feature is particularly useful for finding out who moved the asset -- who moved, when and where -- to pull out the past 72 hours.
Life cycle and inventory
The entire process of entering assets into the bank and ending them should be marked in the system, including maintenance records, depreciation, transfer, end-of-life approval. The inventory function is based on RFID 's profit item: Handheld terminals go over the shelf, automatically form two tables for read-and-not read-out assets, and the difference is highlighted. The machine processing plant mentioned above, which runs from three days to two hours, is this step.
Three areas where hardware and software are highly impervious.
The architecture is so beautiful that the site cannot run well, mostly without a good team. Several common problems need to be avoided in advance.
Clock Sync
All readers, gateways, servers must go to the same NTP (Network Time Protocol) server, with time deviations within 1 second. Otherwise, the same label is read by multiple readers, events are not punctuated and moving tracks are confused. In practice, we have seen the weirdness of a trajectories that occur because the clock does not synchronize, and we've been looking for the reasons for half a day.
Reader power matches software policy
Reader power is too high, neighbouring areas are colliding, software filtering can't keep noise; it's too low, and there's a leak. The general approach is to move the base power from the field environment, then adjust the software to the historical read card success rate dynamic. The values are measured on-site, without all-power parameters.
The abnormally offline can't throw data away.
Network power outages are normal. Both intermediate and handheld terminals must have local caches, where the data read and written during the breakup is stored locally, and then restored to the backstage. Shanghai-based Chosen-Jei usually has intermediates for 1-7 days of local storage, with a specific capacity measured by asset size and frequency reading, and without losing critical events when the network is shaking.
Answer these three questions before you land.
The architecture is designed for business services, and it is best to think three things before going on: first, are assets much more moving or quiet? Much moving (e.g. tools, diems) requires a focus on positioning and trajectories; much quieter (e.g. room servers) focuses on billing and access registration. Second, do the count take time of month or year? High frequency counts require high number of terminals and real-time system performance by opponents. Third, should existing ERP (enterprise resource planning systems) or financial systems be connected? If you need to pick up, intermediates and applications must preset standard interfaces.
Think of all three things, hardware configuration and software are in place to avoid a situation where it costs a lot of money to buy a bunch of readers that can't be used. The RFID is not the more complex and the more business you get, the more you get, the more you get, the more you get.

Ms. Lei
Mr. Pang