IoT systems are excellent at collecting information from the physical world.
Sensors can detect temperatures, equipment status, asset location, movements, environmental conditions, and other occurrences.
But often, the question is then, what do we do with this information?
That’s when AI + IoT (often abbreviated to AIoT) comes in handy, as it combines connected systems, data pipelines, analytics, and AI/ML to enable insights into patterns and processes.
In this note, we’ll describe a simple AIoT architecture that supports the delivery of insights and actions based on connected systems.
Physical World AIoT Architecture
A simple view of an AIoT architecture would include:
Physical World
Identification/Sensors
Connectivity
Data Pipeline
Storage/Processing
AI/ML
Decision
Physical Action
The AIoT stack typically consists of the following key layers, each with its own set of requirements:
1. Physical systems
These are the systems that host the data that we want to observe, understand, and act on.
Depending on the use-case, this layer can be manufacturing equipment, vehicles, warehouses, worksites, facilities, energy infrastructure, mining equipment, and more.
2. Sensors and identification
Sensors and identification technologies collect information from either physical objects or the environment.
Depending on the use-case, the information can be about the object’s properties, location, movement, utilization, or other factors.
3. Connectivity
The information from sensors and identification tools needs to be sent somewhere for further processing.
This is why connectivity is an essential part of any AIoT architecture.
Depending on the requirements, the connectivity solutions can vary significantly.
4. Data pipeline
Most often, the information provided by sensors is not ready for immediate AI/ML processing.
It must go through a series of data preparation steps, including validation, normalization, enrichment, timestamping, aggregation, and more.
It is in this layer that many seemingly simple IoT initiatives can become significantly complex.
5. AI and machine learning
Once the data pipeline provides information in a format suitable for analysis, the layer can start providing insights.
Depending on the requirements, the approaches used for analysis can vary.
This is an overview of what an AIoT architecture looks like.
Now, let’s take a closer look at the data and intelligence layers.
IoT Tells You What Happened
Let’s imagine an example of a connected manufacturing system.
An IoT platform would allow you to collect such data as:
10:31 – Machine A was stopped
10:34 – Material B was received
10:40 – Asset C was moved
10:52 – Production was resumed
This will give you basic insights into what happened, when, and where.
However, it may be hard to determine the reason for the production stoppage.
It could be related to the events described above, but it could also be something else.
It may require additional investigation to understand if there is a correlation, causation, or no relation at all.
This is where analytics and AI can help.
Instead of analyzing individual events, you can look at patterns that span multiple observations.
AI Helps You See Beyond
AIoT systems can provide such insights as:
Identify equipment malfunctions or patterns
Recognize anomalous equipment behavior
Explain patterns or causes in asset movements
Predict production disruptions
Detect bottlenecks
Discover relationships between equipment and production patterns
The key point is that AIoT adds another layer of interpretation on top of the information provided by IoT systems.
In many cases, it’s helpful to think about IoT and AI as a system that follows the “observe-interpret-act” pattern.
To summarize, the core value of an AIoT system comes from its ability to provide higher-level insights on top of the information provided by IoT systems.
Your Data Is Probably Better Than Your Model
One of the most common mistakes that companies make when designing AIoT systems is to focus too much on the ML model while neglecting the importance of data.
Let’s say that you built an innovative anomaly detection model.
However, if your sensors are faulty, the timestamps are incorrect, the device IDs are random, the network has significant delays, and so on, your ML model will fail to provide reliable results.
A good rule of thumb when building AIoT systems is to ask yourself such questions as:
Do the sensors capture the required information?
Is the data reliable, consistent, and complete?
Are timestamps trustworthy?
Can the data and devices be integrated?
How will the system handle missing or incorrect information?
Are devices discoverable and uniquely identifiable?
Is there historical data available?
Who will be able to act on the information provided by the system?
The AI layer is only as good as the supporting systems that allow it to process and interpret the data correctly.
Focus on the Problem, Not the Solution
When designing AIoT systems, it’s helpful to always start with the end in mind.
More specifically, when designing such a system, always start with the question of what you want to achieve.
It’s helpful to use the “problem-driven” approach rather than the “solution-driven” approach.
For example, instead of asking the question, “Where can I use AI?” ask yourself, “What operational decision would benefit from having more information?” or “What do I want the system to be able to detect?”
Once you have a general idea of the problem that you want to solve, it will become much easier to define the requirements for the system.
The architecture will be determined by the use case.
AIoT Across Different Industries
The concepts described above apply to most industries that want to adopt AIoT systems.
For example, manufacturing facilities can leverage AIoT systems to analyze equipment telemetry, production data, asset movements, and other information to identify patterns.
Logistics companies can use connected assets and vehicles to track inventory movements within warehouses, worksites, and transportation networks.
Transportation companies can leverage connected vehicle data to optimize operations and analyze patterns.
Construction firms can use connected systems to track equipment usage, material inventories, and worker activities to make worksites safer and more productive.
Mining and energy companies can leverage AIoT systems to analyze patterns within their large-scale worksites.
The specific requirements for each system will differ depending on the industry, but the general principles of designing such systems will remain similar.
Edge or Cloud?
Another critical design decision that you will have to make when designing an AIoT system concerns where to process the data.
Some AIoT systems will benefit from being able to process data closer to the source, while others will rely on the power of the cloud.
The decision will be based on a number of variables, including latency, connectivity, volume, computational power, cost, security, and other factors.
When designing an AIoT system, it’s critical to carefully analyze the requirements of each layer and select the architecture based on your needs.
There is no one-size-fits-all solution.
Security Should Be Built In
When building systems that connect the physical and digital worlds, security should be a top priority.
An AIoT system can include a number of security-critical elements, including connected devices, networks, gateways, APIs, cloud infrastructure, databases, and applications.
With every additional layer, the attack surface of the system increases, which means that you need to take security into account at every stage.
This also applies to identity management, access control, and other security-focused features.
From Visibility to Insights to Action
One of the biggest advantages of AIoT systems is their ability to provide valuable insights that lead to action.
A simple way to think about the benefits of an AIoT system is to imagine it as a system that follows this pattern:
Visibility
Intelligence
Decision
Action
For example, detecting equipment patterns is useful, but being able to determine that the patterns represent an equipment malfunction is even more valuable.
Being able to notify the maintenance team about the malfunction is even more important.
This is why it’s critical to design an AIoT system as an end-to-end solution that creates value at every stage.
Conclusion
AIoT combines several different fields, including IoT, networking, data engineering, cloud infrastructure, analytics, machine learning, and software development.
The challenge of building such a system is not to develop every individual layer but to make sure that they can work together to solve a specific problem.
Companies that operate at the intersection of AI and IoT are building a number of innovative solutions for the physical world. Aperture Venture Studio is a venture-building organization that specializes in building AI and IoT-driven companies within the physical systems space.
Final Remarks
AIoT is not simply about connecting sensors to AI.
It’s about building an end-to-end system that encompasses the following elements:
Physical Asset
Data collection
Connectivity
Reliable data
AI/ML
Operational insight
Decision
Action
A good system will always be driven by a clear objective.
It will then work backward to determine what data, infrastructure, and intelligence it needs to be able to achieve its goals.
IoT systems will provide visibility into the physical world, while analytics and AI can help transform this visibility into valuable insights.
Operational systems, in turn, can then use these insights to make decisions and take action.
This is why it’s so important to think about AIoT as a system that encompasses these capabilities.
Now, what patterns have you seen as the most valuable ones for building systems that combine AI and IoT?
Top comments (0)