Future-Proof Industrial Applications
TIMELY, ACCURATE, AND ACTIONABLE DATA
Unlock Industrial Transformation
Whether monitoring machine metrics in a factory or customer traffic patterns in a retail environment, information must be received quickly and securely for it to be actionable. Sending data to the cloud increases latency, exposes privacy issues, adds cost, and may result in information being received too late for it to be useful. However, with edge processing, you don’t need to store all the data or send it to the cloud for analysis. Instead, you can locally analyze data that matters and make actionable decisions in real-time. Efficient edge computing reduces the amount of data transfer needed, saving time and money. OEMs and chipmakers are adding Artificial Intelligence (AI) capabilities to their products to capitalize on the advantages of edge processing.
Neural Network Diversity
Industrial needs for AI vary greatly depending on the application. Neural Processing Units (NPUs) may need to perform at tens of TOPS or more. They also may need the ability to run multiple unique Neural Networks (NN) efficiently—something not typically seen in other markets. While one machine may require a single NN with a small resolution (for example, ResNet50 at 224 x 224), large-scale security installations may require multiple concurrent networks at much higher resolutions (for example, ResNext at 1920 x 1080 x 3 and EfficientNet V2 1632 x 2592 x 3). The long lifespan of industrial devices and semiconductors also introduces the unknown to NPUs, in that future neural networks that may not exist today will have to be supported.
AI in Industrial Devices - Use Cases
Industrial use cases for artificial intelligence include:
- Inspection for visual defects, manufacturing predictability over time, or unsupervised learning models in a factory/manufacturing settings
- Examination of scenes for abnormal behavior or unusual traffic patterns in a secured, retail, or industrial environment
- Video feed storage compression: removal of non-important video frames for archiving efficiency
Best AI for Industrial Uses
Future industrial systems will process ever larger neural networks from a growing number of input streams. Processing at the edge will require highly efficient hardware solutions. Adding the Expedera Origin™ E1 or E2 NPUs to your silicon solution can improve system performance without increasing system costs. Expedera offers additional performance without modifying the trained model, ensuring the most accurate results while saving engineering time often spent reoptimizing models. Additionally, Expedera offers capabilities for future NN support to ensure a future-proof product.
Subscribe to our News
Sign up today and receive helpful resources delivered directly to your inbox.