When people talk about AIoT, it is easy to read it as “AI plus IoT,” two buzzwords side by side. In actual projects, Guangzhou Tongwei Smart Technology Co., Ltd. (Tongwei Smart) does not read it as side by side. It reads it as layers: IoT is the senses, and AI is the judgment. Judgment without the senses can only process second-hand information that someone else has passed over.
This division of work decides how the spending should be ordered.
What the sensing layer has to solve is much more than “being able to see the numbers”
A device connected to a smart IoT gateway, with temperature and current appearing on the screen, is only the entry line of the sensing layer. Several further things have to hold before the data counts as usable:
Collection has to be continuous. Data whose sampling interval is unstable will distort a trend analysis.
An outage has to be covered. If the data during a network outage is lost, the operating condition of that period is permanently blank, and faults often occur precisely when the environment is unstable.
Time has to be aligned. If the clocks of several devices are not synchronised with each other, the order of events cannot be judged, and causal analysis has nowhere to start.
Data has to carry an identity. A value without an equipment number, a location, and an operating-condition tag is only an isolated number. Even after it enters the store, it cannot be used.
All of these happen at the sensing layer. None of them can be repaired by a model. Information lost at the collection step cannot be restored by any later algorithm.
A common mismatch in the budget
The budget split of many enterprises’ AI projects is reversed: the large share goes to the platform, computing power, and the model, and the sensing layer is patched together with the cheapest scheme, on the view that “being able to see the data is enough.”
The result is a large sum spent on a judgment capability, paired with a short-sighted pair of eyes. The platform interface is quite complete, and the charts are also attractive, but the underlying data lacks samples, lacks consistent timestamps, and lacks context. Only when the time comes to do some real analysis with this data does it turn out that the work has to be redone from the collection end.
The cost of this kind of rework is usually higher than the cost of getting the sensing layer right at the start.
How Tongwei Smart orders the layers
We split device-edge-cloud as one into four layers. The two business lines share the same foundation:
Device: field objects — equipment and control cabinets, instruments and sensors, environment and energy systems.
Edge: edge access — the SG series rack-mount edge gateways, the BEC series companion edge gateways, and the SDS and IOC expansion modules, responsible for protocol conversion, offline buffering, and local interlock.
Cloud: platform and data — equipment and data governance, alarm diagnosis and work orders, inspection, reports, and permissions.
Use: coordinated reach — PC and large screens, APP and mini programs, API and third-party systems.
The access side supports Modbus, RS485/RS232, SNMP, MQTT, DI/DO, LORA, 4G, and instruments and third-party API. The specific protocol and equipment model still have to be verified item by item. Whether access is possible is assessed against the equipment list, the protocol, and the point table.
AI capability sits in two places on this foundation: the edge side is responsible for real-time judgment and execution, and the centre side is responsible for overall analysis and continued optimisation. The reason for splitting it this way is explained in detail in the next article — some judgments have to be made in the field and cannot wait to go up to the cloud.
For a standard product supply, the customer takes only “device + edge,” and uses its own platform. For a solution supply, the four layers are delivered together. The same protocol stack and data model avoid repeated development.
The place of the senses has value in itself
What needs to be stated is that Tongwei Smart does not make models, and does not promise to build a smart factory for the customer. We design in-house the smart IoT gateway, expansion modules, industry terminals, and platform software. What we do is turn dispersed equipment, environment, and energy sites into objects that can be connected, diagnosed, and operated.
This place does not need to be elevated. The truly hard part of industrial-grade AIoT is precisely making the signals of the physical world become data reliably, continuously, and with context. When that is done well, the judgment layer that follows has room to work.
Next step
If you are selecting for a batch of equipment or for one scenario, you can first send the equipment list and the protocol types to Tongwei Smart. According to the field situation, we give an access-feasibility judgment and a selection list, and then decide whether to apply for a sample test.