How to build an online asset management system? Sensory level access and data processing
RFID Knowledge

How to build an online asset management system? Sensory level access and data processing

PublishedShanghai Zejie Smart Technology Co., Ltd.

The core of the networked asset management system is to get sense layers solidified and then move data. The sensor layer selects sensors and communication protocols by asset type, the edge gateway is initially cleaned and aggregated, the back-stage is used for time series data inventory historical trajectory, relationships to data inventory asset account, so that collection, transmission, storage, visualization can form a complete link. Once the link runs, the count, position, and preservation is dependent on data.

What is the overall architecture of the Web-based asset management system?

A typical architecture has four layers: sensing, network, platform, and application. The sensing layer collects data using devices such as RFID readers, temperature and humidity sensors, GPS/BeiDou modules, and accelerometers. The network layer transmits information over Wi-Fi, 4G/5G, LoRa, NB-IoT, or an industrial bus as appropriate. The platform aggregates, cleans, and stores data. It typically uses a time-series database such as TDengine or InfluxDB for location history and a relational database for the asset register. The application layer supports workflows such as asset counts, location tracking, maintenance planning, and retirement alerts.

Each level of responsibility is clearly divided.

The common error is that all logic is stacked on the application level, and data is stuck. The right way to do it is to have the edge gateway take a part of pre-processing -- filtering, re-profiling, time stamping, initial aggregation indicators, only receiving structured data at the platform level. The sensory layer just "see" and "hear," without making business judgements. We've seen the reverse in the project: sensors are local direct judgment alarms, rules change thousands of solids at a time, and transport is not moving.

What hardware and protocols are to be selected for sensory level access?

Choose hardware based on assets and site conditions. Tag fixed assets and use UHF readers to cut a manual count from several hours to ten or twenty minutes. Equip mobile assets such as forklifts and totes with GPS/BeiDou and accelerometers for meter-level positioning. For precision equipment in cold chains, use combined temperature, humidity, and vibration sensors, typically reporting once a minute.

Communication protocols must balance power use, range, and bandwidth. Use BLE or Zigbee for short-range, low-power applications such as office IT equipment. Use NB-IoT for wide-area applications with small data volumes, such as fire protection equipment spread across a city. Factory equipment commonly uses Modbus or OPC UA and needs an edge gateway for protocol conversion. Wi-Fi suits applications with high data volumes and readily available power, such as electronic tags on warehouse shelves.

The protocol conversion gateway is key to the transition period

Many of the plants' old equipment is used with 485 lines or Profinet, which cannot be replaced overnight. The usual practice is to convert these industrial agreements into MQTT or HTTP and re-enter the platform using a border gateway. The gateway itself needs to support local caches -- data are stored while the network is shaking, restored and then re-routed before they are lost. The MQTT protocol (a lightweight Internet transmission protocol) is more appropriate in this scenario than the HTTP, because it is more friendly to the network ' s unstable environment.

How can different types of assets be managed uniformly?

The asset classification is the first step. It is recommended to divide into two dimensions: high-value fixed equipment (machinebed, server cabinet), high-value mobile equipment (vehicles, instruments), low-value consumables (tools, office supplies). Different categories use different sensory programmes and count frequencies, such as high-value fixed equipment with electronic tags to match door-to-door recognition, mobile devices to real time location, and low-value consumables are only recorded when entering or leaving the vault.

Data models are designed in advance. The master data contains code, name, category, department, date of purchase, depreciation information, base fields such as original values, extension fields are customised by industry, for example, medical devices with "use room" and "sterilisation records", IT equipment with "MAC address" and "configuration specifications". The code suggests that the two-dimensional code or RFID electronic label be used as a physical identifier to avoid a sector-wide set of codes using the standard code within the enterprise.

What happens when we get the data up?

Data processing is in three steps: cleaning, standardization and writing.

Cleaning is mainly about weighting and correcting. When sensors report high frequency, the same asset can generate dozens of duplicates or approximations in one minute, which requires a time window to aggregate, for example, taking the last position per minute. The error is removing obvious anomalies, such as GPS drifting to data points a few kilometres away.

Standardization is the alignment of data from different sources to a uniform model. For example, the temperature-moisture sensor sampling of different models varies at different intervals and needs to be completed to a unified time grid; asset coding rules for different devices differ and need to map to the master data code.

Writes are to be divided according to the type of data. Time series data (location, temperature, humidity) write time series databases, relationship data (assets information, preservation records) write relationships databases. The middle is to buffer by a message queue (e.g. Kafka), and the platform is to match consumption rates to avoid data backlogs. Data from individual items are usually written at peaks in the central library every morning, planning for the capacity of the database and queue to swallow up at this peak.

The rule engine is the key to data value.

The original data collected are not directly visible to the operator, and a rule engine is required. For example, "A 3-hour unplaced move triggers idle warning" "Stemper moistness exceeds the threshold 5-minute triggers alarm." The rules can be visualized by drag and drag configuration, allowing transporters to adjust their own thresholds without changing codes at any time.

What about asset identification and real-time positioning?

Choose identification based on the inventory method. Bulk counts use UHF RFID (860–960 MHz), which reads hundreds of tags at once from 3–5 m and suits warehouses and server rooms. For individual items, use HF RFID (13.56 MHz) or NFC with a handheld for precise identification. For outdoor moving assets, prefer dual-mode GPS/BeiDou. Indoors, where satellite signals are weak, UWB ultra-wideband (within 30 cm) or Bluetooth AoA positioning (1–3 m) is more common.

Positioning data must account for drift and dropped fixes. GPS drifts badly among dense metal racks. Combining accelerometer data and RFID beacons for correction improves stability. When assets move between indoor and outdoor areas, fuse location data to prevent dropping the fix. In one manufacturing project, a forklift's location disappeared for three to five minutes between the workshop and warehouse entrance because there was no handover logic.

How do you keep the system running when it's online?

The connection is not the end point. The sensor system has a failure rate, the sensors for battery power need to be replaced regularly, the RFID label may fall off, and the positioning module will age. It is recommended that a device health board be set up to monitor the frequency of reporting per sensor terminal, battery power, signal strength, etc., below the threshold, to maintain an automatic unit maintenance.

Data quality is also tracked. Statistics on the data integrity of each asset type (e.g., "the proportion of forklifts having a 24-hour trajectory per day"), whether the check below target is a problem with equipment or network. A monthly calibration of data is done to compare the location and actual location of the system's records, and to adjust the positioning algorithm parameters in a timely manner.

The business sector is involved in the iterative. Asset classification, protection rules, alarm thresholds are not always fixed, and will be adjusted for actual usage after a certain period of operation. Shanghai-based Best Technology has deposited a reusable data model and rule template in multiple industry asset-coast projects that can help clients shorten the time they deploy to an effective cycle.

Extensions are planned ahead.

The boundaries of the asset management system will be widened. It may start with only equipment, then tools, spare parts, and people. It is suggested that the platform level use microservice structures (disaggregating the system into a number of separate small services, each responsible for one function), assets, positioning, alerting, statements, individual deployments, new operations without changing the core link.

In an IoT asset management system, technology selection and architecture are 60% of the work. Data governance and business rules make up the other 40%. Connecting the sensing layer provides the foundation for the applications above it. Getting the sensing layer and data flow working before building visualizations and business features has proven to be a reliable approach in most projects.