Internet of Things (IoT) systems are fundamentally positioned to answer a single question:
What is happening in the physical world?
Sensors can be used to capture information regarding the condition of equipment or assets, their location and movement, their environment, utilization, and a variety of other factors. But what good is that information?
What makes sense from an engineering perspective is to take that data and turn it into something useful.
This is where the combination of connected devices and AI comes into play.
By integrating IoT systems and connected devices into an AI/ML architecture, organizations can take that flow of information and make sense of it.
A Connected Intelligence Architecture
A high-level diagram of an AIoT architecture might look like this:
Physical Asset
↓
Sensor
↓
Connectivity
↓
Data Pipeline
↓
AI / ML
↓
Insight
↓
Decision
↓
Action
Each element represents a logical layer within a broader architecture.
Sensors capture information from the physical world or some type of measurement.
Connectivity enables the movement of data between the sensors, devices, gateways, apps, and data platforms within this architecture.
Next, a data pipeline might be used to organize the information for processing.
Then, AI/ML methods can be used for pattern recognition, anomaly detection or other forms of analysis in order to extract some type of meaningful insight.
Decisions can be made based on the resulting information, and those decisions can then be executed as some form of action.
Notice that what we're really talking about is a chain of value.
It is not enough to have a connected device plus an AI method. There are several other elements that need to be incorporated before value can actually be created.
IoT: The Data Infrastructure
There are a variety of things that IoT systems can capture, for example:
Asset location
Movement of equipment
Equipment conditions
Environmental measurements
Utilization
Operational events
But the usefulness of information is highly dependent on issues related to data quality.
For example, it does not matter how good your ML algorithms are if you cannot get consistent measurements from a given sensor; similarly, if there are missing values, this can severely impact efforts to find meaningful patterns.
This is where data engineering comes into play, and these are some of the considerations that need to be made before anything else:
Sampling frequency?
Preprocessing at which point?
Handling of missing information?
Identification of devices?
Time stamping of events?
Historical data storage?
Data Quality monitoring?
All of these considerations can have a major impact downstream when it comes to actually analyzing patterns and extracting something of value.
And Now, What's the Value of AI?
With data in hand and a series of analytical possibilities opening up, artificial intelligence has a valuable role to play.
Let's say, for example, that an industrial system is continuously capturing information about the status of some type of equipment.
A typical IoT application might make that information easily accessible; users would be able to learn the current condition of the equipment or its location.
An AIoT application, on the other hand, could analyze that very same data (current plus historical) in order to identify some type of anomalous event.
In short:
IoT: What is happening?
AI: What does it mean?
AIoT: What should we do about it?
There is certainly a variety of different options when it comes to the actual ML techniques and approaches that might be applied.
In this case, however, it is important to think about the operational requirements rather than the analytical possibilities.
On-Premise, in the Cloud, or Both?
It is also worth thinking through the placement of these different methods.
Some tasks are better suited for execution at the edge while others make much more sense in a centralized cloud environment; this choice will be driven largely by considerations related to bandwidth, volume of data, latency, processing power, and other similar constraints.
An alternative hybrid architecture might consist of:
Sensor
↓
Edge Processing
↓
Filtered / Relevant Data
↓
Cloud or Central Platform
↓
AI / Analysis
↓
Operational Processing
It is important to ensure that whatever infrastructure is selected can actually serve the requirements of the application.
Making Insights Valuable and Actionable
One of the biggest challenges with AIoT systems is that there is no value in having an insight if there is no one who can take action.
At least not if the intent is to operationalize something in the real world.
Another engineering consideration, then, is around connecting intelligence with some type of action.
Depending on the application, that might mean visualizing that information on a dashboard, presenting it to some kind of operations team, incorporating it into an existing application or system, or taking some type of automated action.
The ultimate aim is to ensure that there is a feedback loop from sensing through analytics to understanding and finally to some form of action in the physical world.
Focus on the Problem, Not the Tools
One of the most common pitfalls when approaching AIoT is to focus on the tools rather than the problem to be solved.
A better approach would be to start with the actual problem.
For example:
What physical activity requires more visibility?
What information is currently being captured?
What information might be needed?
What kind of decision needs to be made?
Can connected devices and/or sensors provide the information?
What type of analysis would make that information more valuable?
How will that information be incorporated into some form of decision-making process?
These kinds of considerations ensure that an AIoT architecture is not an end in itself but rather a means to an end.
AIoT in Different Sectors and Environments
An architecture like this can be applied to a variety of different environments.
From a manufacturing perspective, this might involve looking at connected devices or some form of production system.
Logistics is another area where AIoT has the potential to create value, with an emphasis on location and assets. These are just a few examples, with construction being another industry that involves a changing set of conditions with regard to equipment, materials, people, and the work site itself. Additional areas of interest might include transportation, energy/utilities, aerospace, automotive, and mining.
The actual implementation will vary from one industry to the next, but the fundamental challenges remain largely the same, namely finding a way to connect useful physical world data with something that can drive decision-making and increase value.
For a different take on AIoT and its applicability to physical-world environments, Aperture Venture Studio takes a venture building approach to AIoT.
Insights, Not Information: The Engineering Challenge
AIoT, at least as it is currently used, is a combination of artificial intelligence and the Internet of Things.
Technically speaking, it might make more sense to refer to AIoT as something that brings together physical systems, sensing, connectivity, data engineering, AI, and operational processes.
The real engineering challenge is not in capturing data, but rather in finding a way to make the information valuable and then connecting that information to some form of decision-making and action in the physical world.
That is when AIoT stops being a system focused on simply gathering information and starts to become something else entirely: an intelligent operations architecture.
Top comments (0)