The journey
This journey begins with M2M (Machine-to-Machine) technology, an innovation that has given devices the ability to send and receive data over a network and paved the way for an increasingly connected and efficient world.
Soon, networks gained more capacity and devices, which were previously limited to a few specific functions, could now sense or act on their surroundings. The rise of the Internet of Things marked a revolution.
With falling equipment costs and the new possibilities brought about by cloud computing, the IoT gained scale, connecting millions of devices to the network.
Devices became powerful tools for capturing and transmitting more and more types of information, such as temperature, pressure, location, movement, sound, or light.
Companies began collecting data from a wide variety of sources (homes, factories, vehicles, cities, and even people with wearable devices) and remotely monitoring or controlling systems.
Much value has been and continues to be created with M2M and IoT.
However, these systems remain passive and require human intervention, either to make any decisions or to transform data into valuable information.
With the rapid growth of data, it is natural that the traditional IoT model will soon no longer be able to meet the demand for intelligent automation in many situations.
This is where AI comes into play and joins the IoT. With this symbiosis, sensors cease to be mere witnesses and become intelligent agents of change.
AIoT (Artificial Intelligence of Things) systems are capable of learning in real time, performing autonomous actions, detecting anomalies and patterns, predicting outcomes and trends, and continuously adapting to changes in the environment.
The architecture
To make AIoT possible, we need an end-to-end (E2E) architecture that is prepared and optimised to define how intelligent IoT systems operate together. From data generation to obtaining useful information.
It is this architecture that integrates physical devices, communication layers, data infrastructure, analytics, protocols, and applications into a unified and scalable ecosystem.
Conceptually, we can divide this architecture into layers:
- Physical: sensors and actuators.
- Connectivity: bringing device information to the application layers.
- Pre-processing: Edge, Fog or Cloud Computing
- Middleware: cloud platforms that aggregate and process information and normalise protocols from various sources.
- Application: tools and control panels that bring all the value to the user.
Physical layer
This is the layer that interacts with the real world, including sensors, actuators, readers, scanners, meters, controllers, cameras, and embedded systems that generate or respond to signals.
Protocols such as Modbus, Zigbee, and BLE are widely used here.
Connectivity layer
The connectivity layer is crucial to the functioning of an AIoT system, as it ensures communication between devices and systems. It includes protocols and technologies such as 5G, LTE, Wi-Fi 6, and LPWAN that facilitate data transmission on-site (LAN) and between other networks (WAN). The goal is to ensure reliable, secure, and efficient transmission by managing bandwidth, latency, and traffic congestion.
The connectivity layer is a central point in AioT systems, as without it, information does not flow.
Pre-processing layer
This is an intermediate layer that cleans, formats and transforms raw data into structured inputs for analysis. Pre-processing can be performed at different times:
· With Edge AI, this work can be done on the device itself, still in the physical layer, where the technology would be integrated.
· It can also be performed in intermediate layers of the network (fog computing), such as gateways or distributed computing.
· Or even in the upper layers, already in the cloud.
These tasks have a considerable impact on the efficiency, speed and security of the entire system, and the choice of topology depends on business needs.
It should be noted that, in some cases, AI can already start working from the beginning of the data journey, using concepts such as TinyML and accelerators such as TPUs (Tensor Processing Units) or NPUs (Neural Processing Units), which brings intelligence even closer to the source.
MidlleWare Layer
We can think of this as the intelligence layer. This is where all the magic of interoperability happens, where data from various sources and protocols is organised and normalised so that it can interact with each other.
This is where AI and analytics really come into play. This can include the implementation of machine learning and deep learning models for tasks such as prediction, classification, anomaly detection, and optimisation. In addition to supporting model updating, federated learning, and reinforcement learning, with the goal of achieving continuous improvement.
Robust platforms on the market, such as AWS IoT Core, Azure IoT Hub, and Google IoT, are some of the best known, but there are many other niche options as well.
Application layer
In this layer, value is delivered to the user. The application layer transforms insights into actions, visualisations, or automated workflows through dashboards, APIs, control interfaces, and corporate systems such as ERP, SCADA, WFM, billing, or CRM.
The user can interact with AIoT using mobile applications, web interfaces, or voice commands.
The main function of the application layer is to transform data into value for strategic decision-making and return on investment.
This entire architecture must be designed to promote modularity and abstraction, allowing new components to be integrated or changed as technology evolves. This approach ensures that the AIoT system is not only reactive, but also intelligent and adaptive.
We will continue this topic in my next article, so don’t miss it.
