AIoT: A Difficult, But Important Marriage of Technologies
September 22, 2023
When it comes to AIoT, some of the terms and techniques you’ll need to familiarize yourself with include Edge computing, machine learning (ML), TinyML, anomaly detection, natural language processing, computer vision, and predictive maintenance.
Most of these technologies will be covered at Renesas’ AI Live virtual conference. But I’ll give you some of what you need to know here. For example, edge computing unfortunately means different things to different people. In some circles, it refers to the “edge computer,” or the conduit from the local IoT system to the cloud. To other people, it refers to the point at which the data is acquired. This is also known as the end point.
For the context of this blog, let’s go with the latter definition. In fact, there are many examples of where the AIoT system doesn't need to be connected to the Cloud, as it has the chops to handle all the processing internally. In some cases, we’re seeing this occur on lower power MCUs, which is a far cry from what was available a few years ago. Even if the cloud is employed, that connection is streamlined thanks to the “pre-processing” that can be done at the edge. Another relatively newer concept is that this AI work doesn’t rely on special hardware blocks (like TensorFlow processors) and makes great use of the DSP extensions available in Arm M-core devices (M4 and above).
Whether the data is processed at the edge or in the cloud, developers are learning how to take advantage of today’s faster 5G and other low-latency networks. The rollout of 5G networks enables faster and more reliable communication between IoT devices, which is essential for real-time AI processing.
Optimized For Resource-Constrained Devices
The latest AI models, especially deep learning models, are being optimized for deployment on resource-constrained IoT devices. Pared down further, into machine-learning algorithms, the software takes inputs directly from the sensors at the edge and incorporates the data that it needs to optimize its operation. This is key for predictive-maintenance applications, where AIoT is used to predict when IoT devices, or equipment, will require maintenance by analyzing sensor data for signs of wear and tear, reducing downtime and costs.That’s also where TinyML can be designed in.
By deploying lightweight (tiny) machine learning models, the code can be consumed far more efficiently by lower power and potentially less expensive MCUs. These models would obviously be highly optimized for these resource-constrained environments. Natural-language processing comes into play to allow equipment to recognize voice commands or text-based interactions with devices.
Anomaly Detection Covers Diverse Applications
Anomaly detection is a feature that’s part of machine learning. As you have come to learn, IoT devices tend to generate large amounts of data, and anomaly detection techniques powered by AI can identify unusual patterns or events. In addition to helping the machinery run more efficiently, it can also be used for security purposes.
Computer vision is also seeing diverse use from security applications to quality control. Using the AI algorithms, cameras can handle object recognition, facial recognition, and even gesture control. It can spot anything from a faulty product working its way through an assembly line to people who shouldn’t be present in a particular area.
Since technology continues to evolve at such a fast pace, it's essential for developers to stay up to date with the latest developments in AI and IoT to make informed decisions when implementing these technologies in their projects and applications.