IoT systems can take data from the physical world
Sensors can carry readings from sources that include temperature, vibration, pressure, and movement. Identification solutions offer information about location, equipment status, asset attributes, and proximity. Connectivity technologies such as RFID, BLE, UWB, GPS, and others can provide additional context.
Collecting information is one challenge: it's what makes IoT so compelling and valuable.
But the way this data is applied is more interesting.
Artificial Intelligence is often discussed as a means for analyzing information.
But the combination of AI with IoT (or AIoT) suggests something more interesting - a potential fusion of the two disciplines.
The Basic Pipeline for AIoT Systems
An AIoT pipeline can be viewed as consisting of a series of steps:
Physical World → Sensors → Connectivity → Data → AI → Decision → Action
Each layer performs a different function, and the nature of these operations can vary significantly based on context.
Sensors and related technologies such as identification provide access to information from the physical world. Connectivity brings this information to an analyzeable form in computing systems. Data layers may normalize and aggregate information, potentially from multiple sources. AI and machine learning systems can analyze this data to find connections and patterns. Decisions take this information and apply it to business logic or related rules. And action systems attempt to return the results of these operations to the physical world in some way.
Using Raw IoT Data: Are We Getting Signal Or Noise?
If we consider a machine that generates readings that include vibration data several times per second, a system simply storing and displaying these values provides useful information.
But at some point, a person evaluating such a system will want to know what these numbers mean.
When can signal be extracted from noise? How could these numbers be related to other phenomena of interest?
These are the questions that AI can help answer.
By adding additional data about the same physical events of interest (temperatures, production status, and machine state), an AIoT system can attempt to find correlations between data sources. But in this context, note that AI is not replacing the sensor layer of an IoT solution: rather, it is working with that information to generate additional insight.
Bringing AIoT Concepts To Other Industries
The overall architectural concepts would look similar for other industries.
Logistics requires visibility into transportation and related assets.
Connected construction creates similar opportunities for insight into otherwise inaccessible environments and activities.
Energy and utilities can use sensors, meters, and other equipment to get information about the physical world.
In all of these industries, basic concepts are the same.
Connectivity and related technologies provide the means for access. But the specific sensors, data sources, decisions, and actions used will vary based on requirements and opportunities.
Designing For Data, Not Devices: The Developer's Perspective
The most common mistake in many connected-industry initiatives is the assumption that more data is always more valuable.
More information from the physical world always creates new opportunities for insight, but it also creates considerable complexity.
Designing an AIoT system that delivers meaningful data requires careful thought about each layer of the pipeline.
What physical events are of interest? What sources of data can contribute meaningfully to an understanding of those events? How often should such information be gathered? How can different sources of data be connected to provide a complete picture? What patterns might be worth detecting automatically? What should happen when such events occur?
Understanding how the physical world contributes to business logic is essential in AIoT development and design.
Physical AI: Bringing Connected Intelligence To Real Objects
AIoT applications have an additional advantage: they can be combined with physical AI solutions.
Industries involving connected machines or objects can also use identification, sensing, and AI decision support to create Physical AI systems.
These applications go further than conventional AIoT systems by applying AI-generated decisions directly to the physical world.
This approach produces a solution that looks somewhat like a conventional IoT dashboard but with considerably greater depth of insight and analysis.
For an overview of how identification, sensing, and physical AI can combine into a single system, Aperture Venture Studio's overview of AIoT and Physical AI provides additional technical detail.
The Developer's View Of AIoT: Building The Stack
From a developer's perspective, the true value in AIoT solutions lies in the ability to deliver meaningful insight.
That insight is generated when the right information is delivered from the physical world into a software system.
The process of building such a solution involves understanding several considerations related to data collection and delivery, including sources of information, contextual understanding, data normalization, and AI model requirements.
But those who understand the opportunities in AIoT will also recognize the importance of what happens next.
Physical data provides the signal; AI provides the interpretation; and software connects the two, creating opportunities for additional insight and supporting the operational requirements of an AIoT system.
Top comments (0)