IoT has brought us to the doorstep of connecting physical systems to software.
Sensors can capture information from equipment, vehicles, assets, facilities, and the industrial environments in which they operate, and connect them into a digital space in which this information can be stored and analyzed.
However, while capturing this information is valuable, the more interesting engineering challenge lies in converting this data into intelligence.
AI + IoT or AIoT is the process by which we combine these elements.
A simplified version of this process would look like this:
Physical Asset -> Sensor -> Connectivity -> Pipeline -> AI/ML -> Decision -> Physical Action
Each step requires a particular set of competencies and can create value depending on the combination of factors.
1. The Physical Layer
The physical layer is the environment in which the AIoT system operates.
This could include a manufacturing plant or worksite, transportation systems, or any other industrial physical space in which assets, equipment, and activities take place.
Events occur that can be captured digitally to create value:
Equipment activity
Location or movement
Environmental factors
Inventory activity
Operational events
Not all this information may be relevant to your application, and the value to be captured will largely depend on the use case.
However, any entity that has actionable information can serve as an asset to your AIoT system.
2. Sensors + Connectivity
Sensors enable the connection of physical events and actions to their digital counterparts.
Depending on the application, connected devices can capture information on physical objects, their environment, or other factors.
Connectivity turns these observations into information that can be processed downstream.
However, the ability to collect this information is limited by the ability to connect to these devices.
At this level, the ability to collect reliable information is crucial to the success of the AIoT system, because downstream processing will only be as reliable as the information coming in.
That is why AIoT engineering is not just an AI problem.
It is a data and infrastructure problem as well.
3. The Pipeline
The pipeline connects devices and physical systems to data warehouses, processing systems, analytics tools, and AI/ML platforms.
This is the step that brings the information collected by devices into the intelligence generated by AI systems.
It is important to note that the quality of the information directly impacts the value that can be extracted by AI/ML downstream.
As such, information processing and inclusion in the pipeline should also be considered part of the AIoT architecture.
4. AI + ML
This is the point at which AI/ML adds value to the information coming from IoT devices.
The information generated by IoT systems can be processed by AI systems to generate meaningful insights.
Depending on the use case and the type of information, different approaches may be used to enable pattern detection, anomaly detection, predictions, and more.
Crucially, it is important to define goals for AI/ML within the constraints of the information being processed.
It is not enough to simply apply a model on information simply because it is available.
A more interesting approach is to define what decision this information may be able to inform.
5. From Intelligence to Action
An AIoT system reaches its full potential when its intelligence can be converted into action.
Consider the following two statements:
"An asset moved from Location A to Location B."
"The movement of an asset follows Pattern X."
The second statement contains more information and presents opportunities for further analysis.
The next logical step is to decide what should be done with this information.
This leads us to a more complete view of an AIoT system:
Visibility -> Intelligence -> Action
IoT can generate visibility.
AI can generate intelligence from visibility.
People can use this intelligence to take action.
While these concepts are general, there are specific applications within industrial environments that can benefit from AIoT architectures.
Industrial Applications of AIoT
Asset tracking
Inventory and operations
Workforce monitoring
Access control
Industrial intelligence platforms
There are several key enablers to building an AIoT system.
These include:
Engineering Considerations for AIoT Systems
Data
Can the information collected be trusted?
Can this information be used reliably?
Connectivity
Are the devices consistently connected?
How reliable is the link between physical and digital?
Infrastructure
Can the system handle the required data?
Are there enough resources?
Integration
How can this intelligence be used?
What systems need to be integrated?
Problem definition
Is there a problem that needs to be solved?
Engineering an AIoT system requires considering factors beyond the AI/ML model.
It is crucial to recognize that while the model itself is an important factor, it is not a guarantee of success.
The success of an AIoT system ultimately depends on the system as a whole.
Building Around the Problem
A good approach to engineering an AIoT system is to build around the problem, rather than the other way around.
This approach considers the following steps:
What do I need to understand?
What information do I need?
What systems, sensors, pipelines, and analysis methods do I need to get this information?
What can I do with the information?
By structuring the approach this way, we keep our focus on the value to be delivered.
Aperture Venture Studio helps companies build AIoT systems for the physical world by focusing on industrial applications in combination with AI/ML, IoT infrastructure, data pipelines, and application development. Aperture Venture Studio Website
Final Notes
AIoT is often used interchangeably with the combination of AI + IoT.
We propose an engineering-centric perspective on the term that reflects a pipeline from the physical world to intelligent systems:
Physical World -> Pipeline -> Intelligence -> Decision -> Action
The value proposition of AIoT systems ultimately lies in enabling this end-to-end process around a valuable use case.
For developers and engineers working on systems for the physical world, this means that the focus should not be on the model itself, but rather on the complete ecosystem in which it operates beyond itself.
What do you think is the most challenging part of an AIoT system? Is it sensing, connectivity, pipelines, AI/ML, or something else? #AIoT #IoT #ArtificialIntelligence #MachineLearning #IndustrialIoT #IndustrialAI #DevCommunity
Top comments (0)