转动边缘AI数据输入Real-Time行动

全文内容 Full Text

On-device AI can deliver faster insights, greater autonomy, and less cloud traffic — but only with the right infrastructure. More AI is moving out of centralized data centers and into devices that can interpret data and act on it in real time. By handling more processing locally, these edge systems can sharply reduce latency, improve security and privacy, and address growing concerns about data center energy and water consumption. Some data is already processed locally in consumer, industrial, automotive, and medical applications, but that percentage is expected to grow significantly in the next few years as edge devices become more capable and smaller language models are introduced. “Number one are your personal devices,” said Sathish Balasubramanian, head of product for EDA AI at Siemens EDA . “Those are small. For example, a watch that can track your ECG and so forth. If your heartbeat rhythm is off, you can set it up so that it calls your hospital directly. On the industrial side, for example, a control unit monitors pressure in a boiler, or a control unit monitors a particular assembly line in a given section of an automotive factory. At the very extreme edges, every time a robot is picking up something and doing something, the finger is an edge device.” Vehicles contain multiple sensors and processors that process data locally. “I would say cars in general are not edge devices,” said Balasubramanian. “Edge devices are more like the lidar in the car. Lidar finds out all your obstacles and everything, then it needs to be processed on board the car. Every section of the car is an edge device, whether it’s radar, lidar, the camera, or what is sensing in the camera, which needs to make sense of what the camera is seeing. That whole system is an edge device.” The medical space is also filled with intelligent edge devices. “For example, a robotic surgeon has arms and knows the position to make the cut, with the right pressure,” said Balasubramanian. “Then all the systems need to be monitored and connected so that if the patient’s heart rate goes up, or they are trying to wake up, the human needs to figure out what to do.” Whatever the application, edge AI analytics focus on real-time performance and metrics that address how data is gathered, stored, and used, as well as what actions are taken based on that data. That information is then looped back to improve the process. In effect, edge AI analytics systems act as a bridge between different systems, and between people and hardware. “By placing compute near data sources, edge AI infrastructure enables local inference across telecom nodes, industrial gateways, enterprise servers, and smart building controllers,” said William Chen, product marketing group director for design IP at Cadence . “These systems operate at the point of physical interaction, powering applications from automotive and industrial SoCs to wearables, smart cameras, and consumer endpoints.” That connection between local intelligence and physical systems depends on a steady flow of usable data, collected close to the source, and filtered before it moves farther upstream. Data comes from numerous devices and sensors. “From a data‑gathering perspective, edge systems primarily ingest information from two sources — sensors and human or agent interactions,” Chen explained. “Sensors span a wide range of modalities, including vision via cameras, light, environmental, industrial, and more. They generate data at very high volumes and speeds. Examples include multi‑camera video streams captured over millisecond‑scale intervals — or faster. To make this data tractable, raw sensor outputs must be processed locally to reduce bandwidth and latency requirements before being forwarded upstream, either to a local compute cluster or directly to a cloud‑based AI factory for higher‑level processing and reasoning.” Setting up an edge AI analytics system Several key considerations come into play when deploying an integrate