From the Sensors to the Decision: AI and IoT Collaboration

IoT systems make it possible to collect information from sensors, machinery, vehicles, and other physical devices. However, obtaining information is often not enough.

We also need to analyze the data and understand what to do with it.

This is where AI can enter the picture in an IoT architecture:

Physical Environment → Sensors → Connectivity → Data Processing → AI/ML → Decision → Action

1. Physical Environment

An IoT system always starts with something happening in the physical environment. For instance:

A machine is operating on a factory floor

A vehicle is entering a warehouse

Vibration is detected on equipment

Temperature changes in a storage room

An asset is moving towards a specific location

Sensors are instruments that collect measurements about the physical environment. The type of this data depends on the activity at hand.

We should also note that the quality of this data has a direct impact on how effective an AI model will be. You cannot fix wrong information with an AI model.

1. The Data Layer

Depending on the application case, there will be different types of sensors.

For instance, we may collect the following data:

Temperature

Pressure

Vibration

Location

Motion

Humidity

Sound

Electricity

Machine conditions

The data may come in a continuous stream or have a regular interval in between.

At this point, it is important to note that before we can take any action, an IoT system only observes the environment. We still need to understand what is going on.

1. The Correct Connectivity Layer

Now, it is time to send the information somewhere. Depending on the application case, there may be different types of connectivity. For instance:

Wi-Fi

Cellular networks

Industrial networks

Gateways

And others

Not all IoT applications are sending all the data from sensors to a central server or cloud. Some systems are designed to make decisions locally, without relying on the internet.

1. Preprocessing the Information

In most cases, raw sensor data needs some preprocessing before it is ready for analysis. We can do several things at this step, such as:

Removing duplicates

Filling empty values

Data noise reduction

Data normalization

Time synchronization

Feature engineering

Detecting normal operating ranges

For example, a vibration sensor may collect a few thousand readings per second. At the same time, a machine-learning model may look at particular patterns in this data and classify it.

The quality of this step has the most significant impact on how effective the machine-learning model will be.

1. The AI Layer

At this point, we have meaningful information about the environment. Now, we want to analyze it and find patterns. At this stage, we can do several things, such as:

Anomaly detection (find values that are significantly different from normal ranges)

Predictive analytics (predict future values based on historical statistics)

Classification (put information in specific categories)

Computer vision (apply computer vision techniques to still images or video)

And many others

The type of analysis depends on the case at hand. A few words about computer vision: we may apply this type of analysis to still images, video, or augmented reality.