IoT systems are proving to be proficient at a singular task: collecting data from the environment.
Sensors can provide temperature, vibrations, pressure, location, motion, equipment statuses, and other physical signals. RFID, GPS, BLE, UWB, industrial connections and other technologies can provide additional context around the assets, equipment, vehicles, and people involved.
However streaming data from sensors into a system isn't valuable intelligence on its own.
The interesting engineering challenge is what happens with the data after it's collected - this is where Artificial Intelligence (AI) and the Internet of Things (IoT) intersect, also called AIoT.
A Simple AIoT Architecture
A simple approach to an AIoT is often something like:
Physical World → Sensors → Connectivity → Data → AI/ML → Decision/Action
At each stage we solve a different set of problems.
Sensors and identification provide information about the physical world. Connectivity transports that information through gateways, networks, brokers, APIs, and more. Data systems are used for storing, normalizing, and combining information into accessible formats. AI/ML can analyze this information to detect patterns, anomalies, links, or predictions. Finally, decisions or actions can be taken by a person, a process, or a machine.
While this seems like an easy stack, there are many places where systems interop that often create headaches.
Why IoT Data Alone Isn't Enough
Let's say we have a machine that is vibrating more than usual.
An IoT system can collect and expose this metric, but an operations team reviewing the issue might ask some or all of the following questions:
Is there an issue with this machine?
Have we seen this change before?
Is this due to abnormal load?
Has this machine had maintenance recently?
Does this appear to align with another metric or event like temperature?
Do I need to investigate further?
With enough historical and contextual information, an AI can help find patterns that lead to a decision.
The value of AI is to provide enough information that a person can make a better business or operational decision.
The Data Integration Challenge
One of the bigger challenges that companies face with implementing AIoT is rarely the application of the AI model itself. Instead many companies have to grapple with systems and legacy data - industrial systems, IoT sensors and gateways, PLCs, MES, ERP, maintenance records, logistics providers, RFID readers, applications, and more.
Different systems might house different elements of the information needed to apply an AI model. Companies are left needing to normalize data, apply consistent timestamps, link physical assets to logical representations, and more.
Quality → Identity → Integration → Context → Analytics
While knowing that a sensor represents "Machine 42" is useful, knowing which production line it's on, what it was making, when it was serviced and what conditions it was under are also key enablers for analytics.
Where Can AIoT Be Used?
An architecture that connects physical-world signals into data and AI models can be applied in many different ways.
In a factory it could be about linking machine data with production and maintenance information. In logistics it could be about tying into vehicles, equipment, and operations to understand more about the performance of a fleet. In construction it could be about building out a system around equipment, materials, personnel, and jobs. In mining, oil, or energy it could be about connecting a monitored and connected device that can report on its use and state.
The approach would be different in each industry, but the engineering principles remain similar.
AIoT Isn't Just An AI Model
One of the biggest misconceptions around AIoT is that it's simply about applying an AI model on top of an IoT database. The reality is that building something useful and reliable often requires trustworthy data at different integration points, proper contextual and historical information, model performance and guardrails, security, human decision-making, and a feedback loop.
AIoT is as much about systems integration and data as it is about the AI/ML models themselves.
For more on some of these concepts around applying AI, IoT, and related technologies with physical systems see AIoT and Physical AI research.
The opportunity with AIoT is to evolve from connected devices into smarter, deeper systems that can understand and act on the world around it at every level from the machine to the customer.
Top comments (0)