IoT enables us to connect machinery, vehicles, equipment, sensors and various other physical objects to software
but
it doesn't necessarily imbue these systems with intelligence.
A connected machine can provide a stream of data from temperature, vibratory, location, status or other sensors, but what do these signals mean and what should happen next?
This is where AIoT - the intersection of Artificial Intelligence and the Internet of Things - comes to the fore from an engineering and software design perspective.
IoT Captures the Signals
An IoT architecture typically involves sensors, gateways, networks, databases, dashboards and applications.
A given system may capture a stream of data about a physical asset, for instance:
Temperature
Vibration
Location
Status
Runtime
Production events
and so on. This data can be sent on to an analytical layer whether at the edge or in the cloud for storage and processing.
At this point, the question we are asking is: "What is happening?".
Adding AI into the mix enables us to take this one step further.
AI Adds Insights
Machine learning and other AI-related disciplines can consume this data and detect patterns and relationships between variables.
Rather than looking at any single parameter in isolation, an AI engine can examine relationships between variables and determine whether any unusual patterns are present.
Sensor data
↓
Temperature + vibration + runtime
↓
Data processing
↓
AI/ML analysis
↓
Pattern/anomaly detection
↓
Insight
The result of this process might not always be an immediate action, but in many cases the analysis will not drive an autonomous response but instead provide additional context for a human operator or a software workflow to act upon.
The key challenge in many AIoT applications is context.
A given reading might be completely unremarkable on one type of equipment but indicative of a problem on another.
This is why many successful AIoT solutions examine a broad set of parameters.
The overall process might be visualized as:
Physical World → Sensors → Connectivity → Data → AI/ML → Decision → Action
With each layer playing a specific role: sensors observe the physical world, connectivity transports the data, data systems capture and organize it, AI/ML analyze patterns, and applications, dashboards, or other software make decisions or provide insight to operators.
Edge and Cloud Processing
Many AIoT architectures also involve a combination of edge and cloud processing.
Some data may require low-latency processing while other tasks may be batched and sent to more powerful cloud-based systems for analysis.
A given architecture may involve multiple steps:
Physical Asset
↓
Sensors / Devices
↓
Edge Gateway
↓
Local Processing
↓
Cloud / Data Platform
↓
AI / ML Models
↓
Application / Workflow
The choice of where and how to process information depends on variables such as latency, bandwidth, processing power and more.
AIoT Involves More Than Just "IoT + AI"
The challenge with AIoT is that the mere addition of an AI model to an IoT stack is not sufficient to produce useful insights - rather, the context of the data and the application must be taken into account.
A given system may need to address questions such as:
Is this data reliable?
What is the context around this measurement?
Is this value abnormal?
How confident are we in this result?
Does a human need to review this?
What system should this data be sent to for processing?
This is particularly important in AIoT solutions aiming to drive decisions or actions in the physical world.
Practical Applications of AIoT
The architecture discussed above can be applied in manufacturing, logistics, transportation, construction, power generation and a host of other industries looking to make sense of the data from their physical systems. The primary interest for engineers and architects lies not in the AI model itself but the entire chain from collecting data from the physical world to analyzing it and taking action upon it.
Sensor → Connectivity → Processing → Analysis → Decision → Action
Those looking to learn more about AIoT and its relationship with Physical AI can find more information on
Aperture Venture Studio's website
which covers technical aspects of Physical AI and the associated applications and markets.
AIoT represents a shift from merely connecting objects in the physical world to a digital representation of data points, to understanding these data points and incorporating them into the decision-making process in the physical world.
Top comments (0)