IoT systems do one thing well:
What is happening in the physical world?
Sensors can detect temperature, vibration, pressure, location, movement, machine status, or other operational signals. But getting that information is only half the challenge.
What it really means
and
what action, if any, should be taken
are the next set of problems to solve.
This is where adding Artificial Intelligence (AI) to the mix can prove to be interesting.
IoT collects data from the physical world; AI can create meaning from that data stream.
IoT
An industrial IoT architecture can contain many ways of capturing information.
For example, an architecture might include RFID tags, GPS devices, BLE beacons, UWB location sensors, temperature, pressure, or vibration sensors, gateways, or edge systems.
Each component can create value by capturing a unique signal or data stream.
But it is often the combination of signals that has meaning.
A temperature reading, for example, may have limited value on its own, but in combination with an asset identifier, system state, and recent activity, it can create a more complete set of information.
AI adds an intelligence layer
AI can create meaning, correlations, and actions that would not otherwise be possible or practical in an industrial IoT architecture.
This ability implies that while an IoT system may involve discrete events, AI can look across a large set of information and recognize patterns, anomalies, or correlations that have value.
A generalized AIoT architecture could look like this: Connect → Collect → Analyze → Understand → Act .
That description probably makes it sound like everything must be fully automated. But that’s not always the case.
In some applications, the benefit of adding AI is to extract greater meaning and insight for a human operator.
Context is important
Let’s return to our previous example about vibration levels on an industrial machine.
That reading, taken by itself, may represent an anomaly. But what does it mean in context?
An AIoT architecture might want to know what type of machine is generating the reading and what that machine is doing when the value appears.
It might look at other telemetry streams to see if temperature, for example, is also out of range.
It might examine maintenance logs to determine if it the machine has recently been serviced.
Or it might recognize other events of interest that could indicate that the normal operating procedure was not followed.
It’s one thing to think about the physical world, or the operational realm, in abstract terms. It’s another to actually implement a system architecture that reflects those ideas.
A generalized version of that architecture might look like this: Physical world → Sensors → Connectivity → Data → AI/ML → Decision → Action.
Each aspect represents a distinct layer or category of technology and design.
Physical world : Systems, machines, vehicles, and devices generate events.
Sensors and identification : Sensors or identification devices create data from physical events.
Connectivity : Connectivity software or devices translate physical data into digital form and route it to appropriate systems.
Data : Systems normalize and store the data in a consistent manner.
AI/ML : Machine learning methods add meaning, context, correlations, or predictions to create insights.
Decision : An informed decision can be made based on the added context and meaning.
Action : The resulting information can then drive a human action or workflow, or it can drive an automated action.
That kind of generalized architecture has benefits because it begins to separate concerns.
Not every architecture needs every element at once, and different systems may be more appropriately designed for specific use cases.
The AIoT approach works across many industries and environments.
Manufacturing facilities can apply it to production systems and equipment. Logistics and transportation networks can use it on fleet vehicles and tracked assets. Construction sites might utilize it with connected equipment and supply chains. Energy production facilities can examine power generation systems and infrastructure.
AIoT has applications for tracking physical assets, monitoring systems and processes, inventory management and logistics, industrial automation analytics, and operational improvement across many other industries.
For the purposes of this introduction, we’ll turn to the perspective of the developer.
AIoT is a systems challenge, not an AI problem, and not an IoT problem. It’s a systems-integration problem.
At a certain point the value proposition becomes about connecting the dots — creating the most value by tying physical signals to contextual information, analytics value, and operational processes.
It’s about the overall system design and how different elements contribute to a common goal: creating value and taking action.
These considerations cut across device connectivity and events, data and information, AI and ML, security and observability, and even the human interface and experience.
It’s always about context, data, and analytics in support of operational improvements, decisions, and actions.
That’s the AIoT story in broad strokes.
To see more about how AIoT connects physical-world data to intelligent systems, take a look at more writing by the experts at Aperture Venture Studio .
Top comments (0)