Cartesiam, released a new version of its NanoEdge AI Studio, a software suite that runs on Windows 10 or Ubuntu and enables users to generate as well as validate machine learning libraries for embedded systems. With this new release, and for the first time ever, the program features optimizations for our Nucleo-F401RE and Nucleo-L432KC development boards. Users can select one, instead of only choosing a generic Cortex architecture, and download a custom NanoEdge AI library. It simplifies the creation of a proof-of-concept that can run training and inference operations on these boards. Validating the Cartesiam libraries on these two boards is also free, and users can expect the addition of more STM32 platforms at some point in the future, thus testifying to the close and fruitful partnership between Cartesiam and ST.
NanoEdge AI Studio, From Data Science to Machine Learning
Last year, the ST blog sat down with Cartesiam to better understand NanoEdge AI, and how its first machine learning application on our STM32G4 blew our minds. Today, large companies looking to benefit from machine learning must hire one or more data scientists to collect a massive amount of data for months, clean them, and create models that developers will then use in tools like STM32Cube.AI, which converts neural networks into optimized code for STM32 MCUs. While this type of approach works well for classification problems, such as anomaly detection, the biggest challenge is to detect situations never seen before.
NanoEdge AI Studio is a tool that enables embedded developers without deep data science expertise to generate NanoEdge AI libraries that adapt to unexpected circumstances. The magic lies in running both the training phase that learns a complex nominal behavior and the inference phase that detects whether there’s a deviation from this nominal behavior. The entire process can thus run on the same STM32 microcontroller with simple commands controlled by the end-user, such as the push of a button. As a result, engineers can customize their system to its local environment, making it more robust and easier to install.
NanoEdge AI Studio, From the Tool to the Custom Library in a Just a Few Steps
Before the advent of NanoEdge AI Studio, engineers interested in the NanoEdge AI library had to contact Cartesiam, go over their hardware configuration, and the behavior to monitor. Cartesiam would then produce a custom library for that particular client. With NanoEdge AI Studio, there’s now a tool that enables developers to customize, generate, and validate their machine learning library. The utility first asks users to select their Cortex-M architecture and the sensor in the system. They then import a file with values describing the typical behavior that the system monitors, such as the data from an accelerometer on a fan or the electrical information of industrial equipment. NanoEdge AI Studio then automatically tests, optimizes, and sorts the best algorithmic combination among more than 500 million possible combinations and produces a customized library that developers can validate using the embedded emulator.
Now that the software offers free options for libraries optimized for the Nucleo-F401RE and Nucleo-L432KC, running a proof-of-concept is even more straightforward. For instance, in Cartesiam’s smart vibration sensor tutorial, which was published before this latest software update, users can employ an ST development board and a small sensor board to capture the normal behavior of a fan. Users then feed the data to NanoEdge AI Studio and obtain a library that they can call in the main loop to run both a minimum number of training cycles previously defined by benchmarks within the new software, before engaging in inference. A free library running on our development board optimizes performance and ensures developers can quickly evaluate their needs. It also shows how the NanoEdge AI libraries can enable predictive maintenance, smart security operations, or any of the applications that currently benefit from it.
