What is IIoT machine connectivity?

It is the stack that carries machine signals into operational systems: sensors and taps capture signals, gateways translate industrial protocols, networks transport the data, a platform joins it to schedule and product context, and workflows act on the result. IIoT is the industrial-grade version of IoT, built for plant environments and protocols.

What is the difference between OPC UA and MQTT?

OPC UA (IEC 62541) is an industrial standard that models data richly: a subscriber can discover tags and their meanings. MQTT is a lightweight publish-subscribe transport that moves messages efficiently but says nothing about their meaning. They are complementary, and many stacks model data with OPC UA semantics and move it over MQTT.

Do I need edge computing for machine connectivity?

For small deployments, often not: gateways plus a platform can carry the load. Edge computing earns its place when data volumes are high, when latency matters, or when you want context added close to the machines, such as filtering vibration data locally and forwarding only features and alarms.

Can IIoT connectivity work over Wi-Fi or cellular?

Yes, and battery-powered sensors commonly use low-power radios like LoRaWAN designed for years of battery life. Wired Ethernet remains preferred for high-rate PLC data, while Wi-Fi, LoRaWAN, and cellular cover mobile assets, remote corners of the plant, and sites without network drops.

Is IIoT machine connectivity secure?

It can be, with standard practice: network segmentation between controls and IT, read-only access to PLCs, certificates and encrypted transport, and gateways acting as the boundary. OPC UA includes security in the standard itself. The risk profile is managed the same way as any other plant network engineering.

How much does an IIoT machine connectivity project cost?

Hardware for a typical line lands in the four to five figure range: tens to hundreds of dollars per current-sensing point, low hundreds to about two thousand dollars per industrial vibration point, and a few hundred to a few thousand per gateway. Software and integration vary more, which is why the deciding factor is the value of the actions the data enables.

IIoT Machine Connectivity, Explained

The stack is five layers deep, and the expensive mistakes all happen at the top and bottom.

IIoT machine connectivity is the stack that moves machine signals into operational systems. Five layers: sensing captures the signal, gateways translate protocols, networks transport data, a platform adds context, and workflows act on it. Sensors and gateways are commodity now; projects succeed or fail in the context and action layers.

The Industrial Internet of Things gets described either as a buzzword or as a parts catalog. Neither helps someone who has to connect a real floor. This post walks the stack layer by layer, names the protocols you will actually meet, and flags where projects stall. It builds on our IIoT overview and pairs with the machine connectivity guide for the step-by-step version.

What are the layers of IIoT machine connectivity?

Every working deployment, whatever the vendor logos, resolves into the same five jobs:

Sense. Capture a physical or electrical signal: a current clamp on a motor feed, a wireless vibration sensor on a bearing, a photo eye counting parts, a read from an existing PLC, or a native data stream from a newer machine.
Translate. An IIoT gateway speaks the machine's protocol on one side (Modbus, a fieldbus, a vendor PLC protocol) and a modern one on the other, typically OPC UA or MQTT. This is where forty years of incompatible equipment becomes one vocabulary.
Transport. Move the data over plant Ethernet, Wi-Fi, LoRaWAN, or cellular to wherever it will be used, with security handled here: segmentation, certificates, and read-only access to controls networks.
Contextualize. Raw values become operational facts by joining them to schedule, product, shift, and asset records. A count of 4,812 means nothing; 4,812 against a target of 6,000 on SKU 114 during a short-staffed shift means everything. Edge computing often handles the first pass, close to the machines.
Act. The layer people forget to budget for: dashboards, alerts, records updated in ERP and quality systems, and workflows that respond to events. Without this layer the other four are an expensive science project.

Layers 1 through 3 are mature and affordable. Layers 4 and 5 decide whether the project pays.

Which protocols matter for machine connectivity?

Three families cover most floors. Modbus is the elder statesman: simple, everywhere, and embedded in decades of drives, meters, and PLCs, which is why nearly every gateway speaks it. OPC UA, standardized as IEC 62541 by the OPC Foundation, is the modern industrial standard: it carries not just values but semantics, so a subscriber can discover what a machine offers and what each tag means; see our OPC UA explainer. MQTT is the lightweight publish-subscribe transport that moves data efficiently to platforms and cloud, often carrying OPC UA-modeled or Sparkplug-structured payloads.

For machine tools specifically, MTConnect (ANSI/MTC1.4) defines a common semantic vocabulary, and a joint OPC UA companion specification bridges the two worlds. The practical takeaway: you do not pick one protocol for the plant. You inherit several and buy translation, which is the gateway's whole job.

The gateway earns its keep by making 1987 and 2022 equipment answer in the same vocabulary.

What does the hardware actually cost?

Ranges from current vendor materials, because exact prices move:

Split-core current sensors for run-state detection: tens to a few hundred dollars per point, installed without cutting a wire, per suppliers like Milesight.
Industrial wireless vibration sensors: battery lives of roughly 3 to 7 years depending on sampling interval, per Emerson and KCF Technologies, with unit costs from low hundreds to roughly one to two thousand dollars at the high end.
Protocol gateways bridging PLCs and Modbus devices to OPC UA and MQTT: typically a few hundred to a few thousand dollars, each covering multiple machines.
The standards themselves are open: OPC UA specifications are published via the OPC Foundation and MTConnect via MTConnect.org.

A useful planning heuristic: hardware for a typical line lands in four to five figures total, which is why the business case is rarely about the sensors and almost always about what happens to the data. Price the payoff side with our ROI calculators.

Why do IIoT connectivity projects stall?

Three patterns account for most of the wreckage. Pilot purgatory: one line gets instrumented beautifully, the dashboard demos well, and nothing scales because the pilot was built as a special project rather than as the first tenant of a shared layer. Data to nowhere: signals land in a historian or a vendor cloud that no operational system reads, so the plant is data-rich and decision-poor; the fix for this failure is the whole subject of machine data to action. Vintage paralysis: teams stall on the oldest machines, waiting for a budget to modernize controls, when retrofit methods could have had those assets reporting in a week, as covered in connecting machines without replacing them.

Notice that none of these are sensor failures. The bottom of the stack is mature. The stall is organizational: no owner for the data once it arrives, and no workflow that changes because of it.

Where should a plant start?

At the constraint, with the cheapest sensing that answers a real question. Instrument the bottleneck line for run state and counts, land the data in the layer you intend to keep, and put one decision on top of it in the first month: a daily downtime review, an alert that reaches the right person, a report that stops being assembled by hand. Resist the urge to instrument everything before proving anything. A stack that pays for itself on one line gets funded for ten; a plant-wide sensing project with no action layer gets remembered as the year IIoT did not work here. The order of operations is the strategy.

How does Harmony AI fit into an IIoT stack?

Harmony AI operates as layers four and five: the operational layer that receives machine signals alongside ERP, quality, spreadsheet, and paperwork data, joins them into one context, and runs dashboards, alerts, and AI workflows on top; see the platform modules. We are deliberately unopinionated about layers one through three: whatever mix of clamps, sensors, gateways, and native protocols fits your floor, the point is that it all lands in one place with one vocabulary.

Deployment is white-glove and in person: our engineers walk the floor, help select the retrofit-first path for each asset, and wire the resulting data into workflows your team actually runs, from downtime capture to automated reporting. The CLS case study shows the pattern across multiple shops. The stack is not the goal. The Tuesday morning where the schedule already knows what ran overnight: that is the goal.