Connecting a sensor to the internet is relatively straightforward.
Making sense of the data is significantly more difficult.
In an industrial setting, various physical devices might produce temperature, vibration, pressure, location, equipment status, production, or environmental data. The engineering challenge is to build a pipeline capable of transforming those signals into usable information.
A useful AIoT architecture can be viewed as six layers:
Physical Assets → Sensors → Connectivity → Data Platform → AI Models → Applications
So, let's have a look at how each layer contributes to the whole.
1. Physical Assets
The pipeline starts in the physical world.
This can include machinery, vehicles, manufacturing equipment, facilities, tools, inventory, and other operational assets.
The critical point here is that AIoT begins with a physical process. The purpose is not just to gather data but to solve a problem.
One should have a good reason for monitoring something.
2. Sensors
Sensors translate physical parameters into a digital stream.
These can be related to temperature, vibration, pressure, location, equipment condition, and others, depending on the purpose of the particular deployment.
Sensors are important because any further analysis will depend on the data that they produce.
Garbage in – garbage out.
3. Connectivity
Next, the challenge is moving data.
Since an industrial environment can include various kinds of networks, protocols, gateways, edge devices, and legacy systems, connectivity will have to be engineered based on the operating conditions rather than being an afterthought.
Sometimes, processing data closer to the source of equipment can minimize latency and reduce unnecessary data exchange.
4. Data Platform
Data from raw sensor streams can require pre-processing.
A data platform can perform such tasks as:
- data ingest,
- storage,
- normalization,
- timestamp handling,
- transformation,
- device identification,
- historical analysis,
and many others.
This is especially critical if data is sourced from several different systems.
5. AI Models
Finally, when the usable data is ready, machine learning can identify trends and patterns in that data.
Based on the problem, the AI system could detect anomalies, classify events, predict future conditions, find the relationships between the variables, or make a prediction.
However, the model must fit the operational problem, not the other way around.
Just because the model is more advanced does not mean it will give better results than a simpler solution.
6. Applications
This is where the intelligence becomes valuable.
AI outputs could be presented through the dashboards, alerts, workflow systems, maintenance tools, operational software, or decision support systems.
And this is where human context becomes vital as well.
AI system could detect some anomaly, but an operator or an engineer would have to determine whether it was relevant for any further actions.
The Full Circle
Therefore, a typical AIoT pipeline will look something like this:
Sense → Connect → Process → Analyze → Understand → Decide → Act
Interestingly enough, an action will not always be the endpoint of the pipeline.
The results of the action could provide new input into the next analysis circle.
Thus, there would be a feedback loop between the physical operations and intelligence.
Begin with the Workflow
In the creation of an AIoT system, there may be a tendency to begin with the newest sensor, database, or machine learning model.
Instead, a better way to start is with the workflow.
Asking:
What decision are we trying to improve?
And working backwards.
What data is needed? Where does the data come from? How often does it need to be collected? Where does it need to be processed? What machine learning model or logic is necessary? Who needs the output? And what happens next after that?
This will help ensure that the design remains tied to a concrete operational goal.
AIoT, in the end, is not about putting AI everywhere.
It is about ensuring a reliable process from the physical world to intelligent output and back again.
The AI provides analysis. The IoT provides context. Software ties them together. Humans provide the decision making.
This is how an AIoT system adds value. Learn more at Aperture Venture Studio.
Top comments (0)