When evaluating an intelligent operations plan, one question is worth asking early: if the network drops, can it still work?
For quite a few plans the answer is no—all judgment logic is in the cloud, and the gateway only passes data through. In normal operation the problem does not show. When a fiber is dug through, or 4G is disturbed by a thunderstorm, the site returns to an out-of-contact state. For industrial-grade AIoT to land on an IoT site, the first mile must be carried by the edge. The reason is not only network reliability.
Three things the edge must take on
Real-time recognition. Local judgment of thresholds, status, trends, and anomaly patterns. Equipment-room temperature over limit, abnormal pump current, and unauthorized access-control opening: the value of these judgments decays quickly with delay, and a round trip to the cloud and back has already missed the window.
Local linkage. Rule execution and on-site linkage that do not depend on the cloud being online. Temperature over limit starts the standby air conditioner directly, leak detection shuts the valve directly, and the audible-visual alarm triggers directly. This class of action must still hold during a network outage. Otherwise the linkage logic is left to chance.
Stable acquisition. Offline caching, resume-on-recovery, and multi-center upload. This item decides whether cloud AI can get complete data—as the previous article said, information lost in the acquisition step cannot be filled back in by any later algorithm.
Taken together, the three things mean the edge decides two outcomes at once: whether the site loses control when contact is lost, and whether the cloud will be able to make a reliable judgment later.
The industry's judgment on this problem
According to public reports, Wu Hequan, an academician of the Chinese Academy of Engineering, said in his keynote at the 2026 Industrial Internet Conference that industry is entering a new stage driven by agents, and he also pointed out two bottlenecks that are not yet solved: on-site industrial control equipment (PLCs and industrial PCs) does not have enough native compute in reserve, which constrains cross-vendor agent interconnection; and there is a foundational conflict between the millisecond-level determinism required by industrial control and the inherent randomness of large-model decisions, with no mature general solution at present.
These two judgments are worth putting into a plan evaluation. They show this: the path of expecting to move all intelligence up to the cloud, or to hand it to a large model, does not currently work on an industrial site. Edge equipment with local processing and deterministic execution is a layer that cannot be bypassed.
It should be stated that this is the academician's public judgment of the industry stage. It is not aimed at any specific vendor's plan.
The center is responsible for a different class of problem
The edge solves for speed and stability. The center solves for breadth and depth:
Cross-equipment and cross-site correlation, and assisted root-cause analysis—common problems that a single site cannot see appear only after the data is brought together.
Earlier prompts on risk, energy efficiency, and maintenance needs—these require long-period, multi-site accumulation of data.
Ongoing accumulation and reuse of rules, models, and handling experience—a rule verified as effective at one site can be issued to other sites of the same type.
The two sides are not in a competitive relationship. They are layers. The higher the timeliness required of a judgment, the more it sits on the edge. The larger the span of data required, the more it sits at the center.
How Tongwei Smart splits these two layers
The smart IoT gateways and expansion modules developed in-house by Guangzhou Tongwei Smart Technology Co., Ltd. take on the edge-side duties. SG series rack-mount industrial gateways have rich interfaces and strong loading and data-processing capacity, use Ethernet as the primary link and 4G as the secondary link, support simultaneous upload to multiple centers, and suit wired sites such as equipment rooms, power distribution rooms, and archives as a multi-site data-acquisition hub. BEC series companion edge gateways are small, support varied networking methods, and have a high surge-protection rating. They suit DIN-rail mounting in a distribution box or installation inside equipment, and IoT pairing for traditional equipment. SDS and IOC series expansion modules provide standard RS485 / Modbus-RTU interfaces, dry-contact and 4-20mA inputs, and relay outputs, and support module-level linkage and alarm triggering.
On the center side, the AIoT platform takes on equipment and data governance, alarm and diagnosis work orders, and inspection, reports, and permissions.
The delivery boundary should be written clearly: at this stage the focus delivered is visual monitoring, active alarms, and an operations closed loop, plus verifiable assisted diagnosis and reports. Predictive maintenance is opened gradually as data accumulates. We do not replace the customer's core controllers or on-site safety systems. Remote-control permissions and the safety policy are confirmed in writing at the project stage.
A few things worth confirming first at selection
Whether data is cached during a network outage, for how long, and how transmission resumes after recovery.
Which linkage rules need to execute locally, and whether they still take effect during a network outage.
Whether multi-center upload is needed—sending data to one's own platform and the customer's platform at the same time. This is common in integration projects.
What the site environment requires for wide temperature, surge protection, and dual power supplies.
Once these items are confirmed, the practical usability of the plan is basically clear. If needed, the equipment list and site conditions can be sent to Tongwei Smart, and we will provide a selection list, or you can apply directly for a sample unit and run one round on the real site.