Film Story

Réalisation d’une vidéo pour STmicroelectronics.

Transcription : Let me walk you through our AI lab. I’ll show you some exciting examples of the applications you can build when you leverage the power of Edge AI and the diversity of STM32 microcontrollers. Here, we’re using machine learning for vibration analysis to detect anomalies on this industrial motor. The system is built around an STM32WB55 microcontroller and 3D accelerometers. The algorithm running on the STM32 will learn when the machine works well and once done, it will detect unusual patterns very accurately. This is an easy way to add predictive maintenance features to any machine and prevent downtime. To create this library, we used our AutoML tool, NanoEdge AI Studio. The resulting code can be deployed on any STM32, starting from C0 series. It requires no expertise in machine learning, and it’s completely free of charge! Our second demo is all about detecting a person’s body posture. Here, we are using a Neural Network model called MoveNet, to identify 13 key points on a person, and use those for pose estimation. It’s a great feature to analyze movement during fitness workouts, or for health and safety applications, for example to detect falls. The system is built around the upcoming STM32MP2, and the images are captured by a 5 megapixels camera. The algorithm is running on the Neural Processing Unit of the STM32MP2. This dedicated hardware accelerator is mandatory for this kind of computer vision application. This use-case has been built using X-LINUX-AI, which provides users with an extensive framework to implement Edge AI on STM32 MPUs. Our last demo is also about computer vision, with a 3D camera stabilizer that tracks people autonomously. This feature is handled by the STM32N6, our forthcoming microcontroller which features a dedicated Edge AI IP called the Neural-ART accelerator. For this specific application, using a microcontroller makes a lot of sense because of their very low power consumption, and cost effectiveness. This smart gimbal embeds a camera that automatically tracks the subject using a person recognition algorithm, combined with a tracking function. We used the STM32Cube.AI SW ecosystem to benchmark, optimize and compile this model for our neural microcontroller. With the Neural-ART accelerator, the STM32N6 achieves unique performance in this range and opens the door to unique applications that were previously intractable or reserved for MPUs and GPUs. So, this gives you a glimpse of the potential of embedding AI in STM32, keeping in mind that all the SW tools provided are free of charge. And now, let’s hear how Pfeiffer Vacuum, an ST customer in the South of France, is using Edge AI to transform their applications.