IoT systems are incredibly effective at collecting information about the physical world.
Sensors can let you know where an asset is, how an operation is performing, how the environment is changing or how often an item is being used. But when an IoT deployment matures, data collection is only part the picture.
What happens next has more to do with how you can use that data to make changes to your operations.
That's where AI + IoT (or AIoT) comes in: using connected devices and their data with AI and machine learning to detect patterns, find anomalies, make predictions and drive operational decisions.
IoT Answers "What Is Happening?"
A typical architecture for IoT looks something like this:
Physical Asset
↓
Sensors
↓
Connectivity
↓
Data Pipeline
↓
Storage
↓
Application / Dashboard
This architecture can be tremendously valuable. It lets you learn more about your physical environment by collecting and analyzing data about assets, equipment, the environment, or utilization. This is a huge leap from what you could do before IoT was involved. This is a system that can give you a much more observable physical environment.
But what does that really mean for someone looking at a dashboard with thousands of data points? How does that turn into actionable insight for an operator?
That's where adding an intelligence layer can be tremendously valuable.
AI Adds Context to Connected Data
AI can consume connected data and detect patterns or anomalies which might not be readily apparent.
The architecture then looks something like this:
Physical Asset
↓
Sensors
↓
Connectivity
↓
Data Pipeline
↓
AI / ML
↓
Decision
↓
Physical Action
Notice the difference? IoT primarily answers the question:
What is happening?
whereas AI can help answer the question:
What does it mean, and what should we do about it?
As a specific example, take equipment data. By analyzing historical data about equipment, you could detect patterns or outliers that indicate a change in the expected performance. From location and movement data you could detect traffic patterns or unexpected movements of physical assets. Environmental data could help add valuable context about changes in the environment.
Again, the point isn't to add AI simply because you're using IoT. But the value of extracting useful intelligence from information you already capture is undeniable.
The Data Pipeline Matters
One of the most common pitfalls when working with AIoT is to think the model is the most important part while ignoring the data plumbing that feeds into it.
In practice, the intelligence you get out of the system is only as good as the data that was fed into it. Developers and technical architects need to consider data-related questions such as:
Are sensors providing trustworthy data?
How often is the data being captured?
Can different sources of data be correlated?
How are exceptions handled?
Can the data pipeline handle the volume of data being processed?
Where does the AI inference take place?
How are model results consumed by downstream applications?
None of these are purely machine learning questions; they're system design questions. No matter how sophisticated your model is, it can't make up for a faulty or inconsistent data pipeline.
From Data to Operational Decisions
This is where the rubber meets the road for an AIoT project. It's when intelligence from AI starts driving physical decisions.
The difference between a system that simply uses sensors and one that uses sensors plus an intelligence layer can be huge. Here are two example systems:
System A:
A sensor detects the value of a condition about some equipment and sends it to a dashboard.
System B:
The system captures data about that equipment over time, finds that a certain value is exhibiting an unexpected pattern, uses that context to determine an appropriate course of action and delivers that as an insight to an operator.
Both systems use sensors, but the second one uses an intelligence layer to turn that sensor data into an actionable insight. This can be summarized as:
Visibility → Intelligence → Action
That progression is a great way to think about how AIoT systems should be architected, because it's outcome-focused rather than technology-focused.
Start With the Physical Problem
AIoT projects can get very complicated, very quickly. You can spend weeks or months talking about sensors, connectivity, cloud platforms, data engineering, machine learning and applications before you even deploy anything. That said, the conversation shouldn't start with the technology.
Start with the physical problem.
What do you want to understand? What information isn't available to you right now? What decision would that information help you make better decisions? What would you do differently with that information? Once you've got that information in hand, it's much easier to determine both what data you need to collect and where AI might be helpful.
This approach is particularly relevant for organizations that are building AIoT ventures around industrial applications. Aperture Venture Studio is one example of a venture building studio focused on AIoT and connected physical systems.
The Bigger Picture
AIoT is more than simply connecting AI to IoT. It truly is about building a connected system in which information about the physical world can flow through increasingly sophisticated processing layers before it feeds back into operations in the physical world.
For developers, that means looking at the system as a whole rather than the individual parts. While each component is important, each one feeds into the system in different ways. By building these systems you can move beyond simple IoT and start using AI to make the most of your operations.
Top comments (0)