IOT solutions have one primary purpose:
Getting information and data on what is happening in the physical world.
Sensors can get information about temperature, vibration, position or movement, pressure, status, and other measures. Getting that information is valuable, but its not' the end.
The other side of the puzzle is making that data valuable for software and people.
It's where AI plus IoT gets interesting.
IOT Collects the Signals
An IOT architecture typically focuses on physical assets and connects them somehow with sensors.
A basic data path would look something like this:
Physical Asset -- Sensor -- Connectivity -- Data Pipeline -- Application
In this scenario the physical asset varies according to need or environment, possibly including a device, machine, car, production process, inventory items and more. These would typically have some manner of sensor attached, which in turn would collect certain type of information:
Temperature
Vibration
Location
Movement
Pressure
Status
Producing or Processing Event
Environmental Conditions
What makes this valuable is that these streams of information aren't directly useful.
A temperature reading is only a fraction of the picture. Its the pattern of collected temperature values, along with other measurements, that can be useful.
AI Adds an Interpretation Layer
This is where artificial intelligence and machine learning come into the picture.
The AIoT architecture can be conceptualized as:
Physical World -- Sensors -- Connectivity -- Data -- AI/ML -- Decision -- Action
The idea here is that an AI can process and interpret patterns.
Using the previous example, a machine is making vibrations that get captured by a sensor and sent to whatever IOT architecture or pipeline exists. An analytics or machine learning system can process that data, find trends or patterns, recognize what is not normal and flag that as an issue. The point here is that AI is only as good as the data it consumes, which in turn is only as valuable as the original use case or intention.
The Data Pipeline Matters
It's easy, when thinking about AIoT, to focus on the AI itself, but this misses everything that goes into getting data to the AI in the first place.
For starters, the system needs to collect data about a particular physical asset or process. This means designing, building, deploying and maintaining pipelines that collect this information. Other considerations include:
Ensuring the devices are connected (or get them connected)
Creating or curating the information stream
Preprocessing the information stream into a useful or consumable form
Storing information for later use and reference
Analyzing with AI or machine learning tools
Creating applications for end users
Taking some type of operational or business action
Each of these plays a part and is important. If the data coming into the AI isn't complete, accurate, or in a usable format, the AI won't be able to do its job.
Another consideration is the balance between edge processing and centralized cloud processing. Depending on circumstances, it may make sense to process information closer to the source (reducing latency and bandwidth needs) compared to sending it to an centralized server or cloud instance for processing. Both approaches have their pros and cons, and the right choice depends on factors such as the nature of the data and applications, overall bandwidth availability, and responsiveness.
In practice, it could look something like this:
Devices --> Edge/Gateway --> Cloud --> AI/Analytics --> Application
All of this combines into a working architecture, although the details will vary based on the application. This is still a work in progress, but the main idea remains the same.
Start With the Problem, Not the Model
When it comes to AIoT initiatives, its easy to lose perspective and get wrapped up in the technology itself at the expense of the larger system or process that the technology supports. With that in mind, it's always best to design a project around the problem first. What does a physical-world problem look like?
What information about the physical world do you have? What do you wish you had?
What particular event triggers some kind of process or change?
What should someone know that they don't already know?
Where should an answer go? Where should someone be notified?
Who should decide what? What should someone do with this information?
The answer in each case is probably something fairly specific. It could be anything from gaining insights or visibility into material positioning and movement processes to detecting or predicting certain types of activity or events within a larger physical system.
From Visibility to Action
The value of AIoT solutions is that they can transform information about the physical world into something useful. That in turn, can be connected to another system, application or process that takes that information and does something with it.
Its the link between what the sensors see and what operators, managers and others need to know or do to get things done. This is a rough conceptualization, but the idea is similar or identical to the following progression:
Connect
Collect
Analyze
Understand
Act
AIoT, with its combination of connected devices and artificial intelligence, will only grow in value and scope. The true opportunities will be for those willing to look beyond the technology itself and focus on how it can be used to address a particular set of issues or solve an operational problem. For more information on AIoT, as well as an overview of this approach, Aperture Venture Studio provides an overview of this approach.
Top comments (0)