IoT devices are good at collecting and capturing data.
Sensors report temperatures, vibrations, and other parameters as physical variables. But, given the volumes of data generated by many IoT endpoints, there's a follow up question that arises:
What can we do with all of this data?
This is the point at which the value proposition of Artificial Intelligence (AI) and the Internet of Things (IoT) intersects.
IoT: Layer Of Data Collection
An IoT foundation is generally understood to sit between physical objects and some means of receiving, storing and processing their data.
A manufacturing context can provide one example, with sensors attached to the equipment in a production space generating information about that machinery's vibration, temperature, status and similar metrics. Others might include:
GPS devices to track the location of vehicles
RFID to mark the presence of objects
BLE to capture proximity of objects in a space
Environmental sensors to measure temperature and humidity
Equipment reporting status and availability
The result is a series of datasets intended to capture information about the physical world.
But how does any of that information translate to useful knowledge about that same physical world?
This is where AI can help.
AI: Finding Patterns And Connections In Data
A key property of AI is the ability to find patterns within a dataset that might otherwise appear as just random noise.
If a machine's vibration levels appear to shift over time, that information in isolation isn't necessarily valuable. But when that same data is cross referenced with the status of that machine, it's potentially a sign of friction. That, in turn, might be informative about the machine's predicted useful life.
In this context, AI doesn't focus on individual datapoints but relationships and the connections between them. It takes a broad approach, one in which every input is examined for correlation.
A basic framework might be:
Connect -> Collect -> Analyze -> Understand -> Act
IoT provides much of the pipeline, while AI focuses on the analysis.
The main problem to be solved in many AIoT applications concerns context.
If a sensor indicates that the temperature surrounding some physical object has gone up, how useful is that information? What does an operations team actually do with that knowledge?
Now factor in some additional information:
What asset generated the reading?
Where is that asset located?
What type of material is it composed of?
What is the expected temperature range for this object?
How long has the temperature been rising?
have similar fluctuations occurred in the past?
With that same reported value, the richness of context is dramatically increased.
This is why many AIoT applications include mechanisms for combining and correlating data as part of their overall foundation. It's not enough to look at one isolated metric.
The Challenge Of Contextual Awareness
A key feature of most AIoT applications is the ability to connect and combine data from different sources.
In practice that tends to mean pulling in information from multiple sensors as well as other systems within an industrial application.
Physical Assets -> Sensors -> Connectivity -> Data Platform -> AI/ML -> Insights -> Action
It's a broad category that encompasses production tracking or predictive maintenance, among other use cases. Varying infrastructure approaches are likely to emerge depending on the particular application space. But the general idea continues the pattern of uniting information from the physical world with methods of analysis and interpretation.
The result can be new applications in predictive maintenance, equipment tracking, inventory management systems, production insights and other areas.
AI + IoT: More Than Just Placing AI On Top Of IoT
At this point, it's perhaps worth emphasizing that building AIoT applications is more than an exercise in adding AI to an existing IoT foundation.
In practice, the more challenging effort tends to be understanding what information is relevant and how to acquire it.
A compelling AIoT application needs a problem to solve, the right set of sensors, dependable connectivity, relevant data and an analysis approach that contributes meaningfully to that original goal. There is value in nuance, not just in volumes of information.
A single sensor or even a collection of them is unlikely to provide substantive insight. The intelligence isn't in what can be measured, but what can be understood based on what has been collected.
With that said, there are plenty of developers and engineers looking at practical applications of AIoT in the wild.
Aperture Venture Studio is one of them, focused on building an array of value centred on industrial applications.
As an idea, AIoT creates the potential to understand and act upon the physical world in new ways.
IoT has made it much easier to digitally represent that world. But how do we go from there to some meaningful set of actions? The role of AI is to begin to make sense of that data and find value and patterns within it.
It's one thing to have a rich view of the current state of the physical world. The compelling aspect of this combination is what we can do about it.
Top comments (0)