Consider the analogy of a factory as a human body. The machinery doing the physical work is similar to the muscles. However they alone possess no knowledge of when to move, rest or when something is about to tear. We need nerves to carry signals from the brain to the muscle. That nervous system of nerves is what we call the Industrial Internet of Things or IIoT.
To be clear, IIoT isn’t a fancy word for sensors, it’s the entire pathway that takes a physical event as an input and turns it into a data point that a computer and eventually a person can act on. It could be anything ranging from a small temperature spike in the machine to a huge vibration in the motor.
You can have the smartest AI model in the world trying to predict when a server would crash but it still won’t work unless there’s an IIoT working with it.
Here’s the problem though, the hardest part isn’t collecting data. Most manufacturers generate tons of data. It’s turning that raw data into something that influences a business decision. Solving the gap between OT and IT is what matters.
To do that, we need to first look into how IIoT functions in layers. Think about those old school handwritten letters in the 90s. You physically write the letter, it ends up at the local post office, gets sorted and travels through the transportation network and finally lands up in the right mailbox. Data in a factory behaves similarly.
Layer 1: Perception, the letter gets written
This is the part where the data is born. Sensors physically interacting with the machine and collecting the data. They also include visual data such as cameras scanning for defects or safety issues.
It should be noted that this process gets messy in practice especially because many of the old factories weren’t designed to be sensor friendly. The sensors need recalibration over time otherwise they keep drifting out. The old parts also generate a lot of noise especially from the nearby motors. This problem of putting new tech over old infrastructure is called the brownfield problem.
Layer 2: Network, the letter goes to the post office
Once the reading is recorded it’s the network layer’s job to move the data to its final destination. This process can be wired or wireless and often uses a mix of technologies such as Wi-Fi, 5G, Zigbee, or LoRaWAN, plus specific communication protocols built for industrial settings .
Layer 3: Edge processing, the local sorting facility
Imagine every letter from every neighborhood going straight to a single headquarters with no sorting . The system would simply collapse. That’s exactly what happens if you stream every data packet to one central server. To prevent that a small computer is installed near the equipment that filters, aggregates and cleans the data before sending it. Only the interesting stuff such as anomaly or a data packet crossing a certain threshold is forwarded. This process is also known as edge processing.
Layer 4: Application, the data becomes a decision
This is the layer where the clean data is used for predicting failures, where digital twins get updated and where energy management tools operate.
Layer 5: Business, the letter reaches someone who can act on it
The final layer connects the IIoT data to the business domain.
ERP software that manage functions such as finance, supply chain and procurement.
Manufacturing Execution System (MES) software that manages what’s happening on the shop floor right now, in real time.
Product Lifecycle Management (PLM) tracks a product’s data from initial design all the way through to retirement.
The protocol duel: OPC UA vs. MQTT
A factory might have hundreds of different machines trying to talk to the same system. How do they agree on the shared language ?
A protocol is like an agreed upon grammar for how devices exchange information. For the industrial domain OPC UA and MQTT protocol dominate the field.
OPC UA (Open Platform Communications Unified Architecture) works similar to you asking a specific question to a librarian. It’s a client server model where one device directly requests information and gets a response with labeled contexts containing info on the data, the units and how it relates to other data. It’s rich in information but lags behind in high computing costs and more setup work. This might be an issue when you are dealing with a mix of old and new equipment.
MQTT (Message Queuing Telemetry Transport) on the other hand functions much like a radio forecast. A device posts messages to the channel and anyone “subscribed” to that channel can pick it up through the middleman known as the broker. It’s light weight and is ideal for thousands of small sensors sending data however the tradeoff here is that the data isn’t rich in context.
Where should the thinking happen: edge or cloud?
Edge computing means the processing is done physically close to the machine as stated earlier. This has several advantages. The machine can detect an anomaly and react in milliseconds without waiting for a trip to a distant server. You are saving bandwidth as the edge filters don’t allow noise to pass through so no risk of clogging the network. Finally if the internet connection drops which is more common in older factories the edge still monitors critical factory equipment instead of the whole system going blind.
Cloud computing, by contrast, offloads heavy processing to massive centralized offsite servers. It excels in a few key areas:
Scale : These systems aggregate years of historical data from thousands of equipment nodes across multiple sites simultaneously.
Heavy computation : Training advanced machine learning architectures, such as LSTMs or Transformers designed to spot temporal trends, demands massive computational power that a small edge device simply cannot provide.
Pattern recognition : An edge node only monitors a single piece of equipment, but central servers analyze metrics across whole fleets to detect widespread failure trends. They can then deploy refined predictive models back down to local units.
Because of this, modern industrial setups rarely choose between the two. They rely on local nodes for real-time reactions and leverage offsite servers for high-level macro analysis.
The next frontier: Green IIoT
Deploying vast networks of continuously operating equipment introduces a fresh challenge: the energy consumption of the hardware itself. The effort to minimize this power draw and environmental footprint is known as Green IIoT.
So what are we doing about this ? Some of the key notable developments :
Energy-aware hardware : Nodes remain in ultra-low-power sleep states until an event triggers transmission, eliminating the constant power draw for listening.
TinyML: Setting up compact machine learning models directly onto low-power microcontrollers which helps in decision-making without transmitting raw data packets offsite.
Sustainable hardware design: Extending component and battery lifespans to decrease e-waste and lower maintenance labor involved in servicing.
Stay tuned for the next part of this series which will focus on Digital Twins !
This framework is synthesized from the landmark systematic reviews on industrial IoT, digital twins, and the digital thread published by Haqiq et al. (2026), Chen et al. (2025), and Abdel-Aty & Negri (2024).
Subscribe
Login
0 Comments
Oldest