Definition
Industrial IoT (IIoT, Industrial IoT) connects industrial devices such as sensors, PLCs, robots, and AGVs through wired/wireless networks, enabling data collection, remote monitoring, predictive maintenance, and production line digitization. Communication technologies include Industrial Ethernet, TSN, NB-IoT, LTE-M, Cat.1, MQTT, AMQP, OPC-UA, etc., which need to meet requirements such as high reliability, low latency, secure isolation, and edge collaboration.
Communication Requirements and Selection
- High reliability, low latency: Control services require uRLLC-level assurance; data collection services can tolerate some delay
- Industrial protocol support: Modbus, OPC-UA, Profinet, etc., with gateways or platforms needing protocol conversion
- Secure isolation: Private network, VPN, Firewall, meeting MLPS (Multi-Level Protection Scheme) and industrial control security
- Edge computing collaboration: Local pre-processing of data to reduce backhaul bandwidth and latency
The Role of Cellular IoT in IIoT
NB-IoT is suitable for static, low-data-rate, deep-coverage sensors; LTE-M is suitable for mobile, voice-capable wearables/vehicles; Cat.1 is suitable for medium-data-rate applications such as mobile payments. Private network APN, fixed IP, and QoS ensure key business services.
Typical Use Cases
Device monitoring, predictive maintenance, energy management, production line digitalization, AGV scheduling, smart warehousing, etc. SimtoLink provides industrial-grade IoT SIM cards for NB-IoT, LTE-M, and Cat.1, combined with private networks and cloud platforms, offering a one-stop connectivity solution for IIoT.
Industrial IoT is a key application area for SimtoLink. SimtoLink provides industrial-grade IoT SIM cards based on NB-IoT, LTE-M, and Cat.1, and achieves secure isolation with dedicated APN and VPDN. The SimtoLink cloud platform supports industrial protocols such as MQTT and OPC-UA, providing one-stop connectivity and data management solutions for scenarios like device monitoring, predictive maintenance, and energy management.
Periodic samples and sudden events should not share one channel
An industrial site has two kinds of traffic at once. One is temperature, current, and output that arrive on a beat. A few hundred milliseconds late can usually still be used. The other is emergency stop, a limit exceeded, and access control. Late, it no longer means anything. Put them in the same queue and a large historical backfill blocks the alarm. Split them into two channels. Periodic values can be merged, compressed, and allowed to be retransmitted. Event values stay short and get a higher send priority. To see whether the split is right, trigger an alarm while history is being backfilled, and see whether the alarm still arrives on time.
If a control loop needs a fixed beat, jitter on a cellular link is often larger than that beat. Leave that loop on site. Cellular should carry status and setpoints. Putting the remote network directly inside a servo cycle means one handover opens a hole in the beat. What can be done remotely is monitoring, parameter download, and analysis after the fact, not replacing the clock of the field bus.
When clocks are not aligned, several curves cannot be lined up
If every device reports with its own power-on timestamp, drawing them together in the cloud shifts the order of events. Time alignment has to be done on the gateway or the device, and the time source has to be named. If there is no way to align clocks, at least stamp the arrival time from one clock at the moment the gateway receives the sample, and keep the device’s local sequence number. Keeping only the arrival time lets transport delay reverse cause and effect. To see whether the data can be aligned, see whether the time difference of the same event on two devices is stable. If it is not stable, fix the clocks before talking about algorithms.
Points with different sample rates cannot be subtracted sample by sample. Aggregate the fast points onto the beat of the slow points, then compare. If a gateway pours an unaggregated high-speed stream straight into a narrowband uplink, it fills the link and the low-speed alarm is delayed with it. Write the aggregation window into the point table. Do not guess it later in the cloud.
Fields dropped in protocol conversion cannot be filled back in the cloud
Point name, quality bit, and engineering unit in a field protocol, once the gateway turns them into a generic message, leave the cloud with a bare number if the mapping table has no columns for them. When the quality bit says the sensor has already failed, the bare number can still be the last normal value. A mapping should at least keep the point name, the value, the time, and the quality. To separate a conversion problem from a collection problem: if a local bus debug tool can see the quality change and the cloud has no matching field, the mapping dropped the information.
The industrial connection SimtoLink provides is mainly the cellular card, private-network isolation, and platform access. It does not replace control logic inside a programmable controller. The card is what connects the gateway safely to the enterprise or the cloud. The private network is what reduces addresses exposed on the public network. The point table and the quality bit still have to be configured in the gateway project. If a value on the platform stays unchanged for a long time and card traffic is also very low, check first whether the gateway is still reading the field bus, then whether the card is online. If both are normal and the quality bit is missing, go back to the mapping table.
Cache during a gateway outage needs a sequence number that does not run backward
When the cellular uplink breaks, an industrial gateway often stores samples locally first. If it backfills them in one burst by sample time after recovery, with no monotonic sequence number, the cloud cannot tell whether the backfill overlaps live data. The same point is written twice, or a new value is overwritten. A cached record needs at least a point name, a sample time, a sequence number, and a quality. The sequence number only increases. The cloud deduplicates on it. Time is only for lining up curves. When the cache fills, the drop policy has to be explicit: dropping the oldest periodic values is acceptable, and dropping an alarm that has not been sent is not. Keep the two kinds of data apart, so an alarm is not pushed out of the buffer by periodic values.
If the backfill rate is higher than the field-bus sample rate, it can still fill the uplink, and a live alarm queues again. Limit the backfill rate, and let alarms cut in line. A SimtoLink card and private network guarantee the path from the gateway to the enterprise side. They do not number the cache. If platform traffic jumps the moment the path recovers, see first whether it is a backfill or a fault storm. A jump with no sequence number cannot be replayed. The whole stretch has to be discarded. During joint commissioning the gateway project should pull the uplink, then confirm that the first record after recovery continues the sequence number from before the break.