How Edge AI is making factories super-brainy
In an era characterized by an explosion of information, where everything is interconnected, we stand at the forefront of the Fourth Industrial Revolution – the golden age of innovation. As the core driving force behind the industrial intelligent transformation, edge AI is rapidly becoming the key to more efficient and intelligent manufacturing processes. In particular, the rise of “small edge AI models” is leading the revolution towards broader application fields, marking the beginning of a new era for devices with independent thinking.
Edge Computing: Endowing Machines with a Smarter and Quicker “Brain”
In a traditional industrial context, data is transferred back to the cloud for analysis and processing before the machine receives the command. This creates a process that is safe but not immediate. Like a phone call to technical support, waiting for the response often requires patience and sometimes causes frustration. What if the production line experiences a disruption? Would an empty production schedule be acceptable?
How do you solve the problem of this delayed response? The most straightforward solution is to upgrade the “brain” of your equipment, allowing it to make the decision on site. Let edge AI work its magic by “relocating” the computing tasks from a remote server to the side of the equipment, thereby enabling real-time data processing and quick responses.
Avnet is rising to this challenge by pushing the frontiers of what has been thought possible. For example, is it possible to help large models unlock their full potential by enabling “smarter” edge computing? Most people are likely to associate large models with large-scale AI systems, such as ChatGPT. This type of AI system requires millions or even tens of millions of samples, powerful algorithms, and a significant amount of energy. What steps should be taken by small- and medium-sized enterprises, or those in application scenarios that require specific product monitoring? Do they truly have to collect tens of thousands of data points from each machine?
By tackling these questions head on, Avnet is dedicated to equipping you with the advanced technology you need for optimal efficiency.
AI Training: No Longer Limited to Big Data
Avnet Silica, the European semiconductor specialist division of Avnet, partnered with Deep Vision Consulting to develop a Defect Visual Inspection (DVI) system that realizes the dream of training AI models at the edge. Unlike traditional AI training methods, the DVI system can perform training and optimization at the edge using a small number of data samples from the production line. The training was limited to the inspections on specific products, much like teaching a child to identify fruit. By showing the child a few different types of fruit, they quickly learn to distinguish between an orange and an apple. This method not only reduces the cost and time required for data collection, but also completes tasks with high efficiency and accuracy.
The DVI system currently runs on platforms based on either an NXP i.MX 8M Plus or NXP i.MX9 application processor, which includes NXP’s neural processing unit (NPU). NPU is a specially optimized hardware unit that can facilitate super-efficient deep learning to support AI training and inferencing.
Predictive maintenance is one of the most high-potential applications for edge AI. Traditional maintenance usually relies on regular inspections or repairs after equipment failures, which results in the inefficient allocation of human and material resources, as well as unnecessary disruption. For instance, by analyzing the condition monitoring data of key equipment, such as the vibration frequency and temperature change of electric machinery and fans, edge AI can forecast possible device failures and provide early alerts. Enterprises are then able to arrange timely repairs and avoid the production line disruption and high repair fees resulting from equipment failures.
Vision technology plays a crucial role in the continuity of online predictive maintenance. Though the image sensor and lens used in the DVI solution mainly operate in the visible part of the spectrum, the design also supports infrared, ultraviolet, and X-ray sensors. The system provides an anomaly score for images of samples inspected, indicating the degree of defectiveness. The user may define the threshold between good and bad or pass and fail to set the sensitivity of the system. Images can also be annotated to show the exact location and classification of the defect.
Low-Power Consumption: Life Extension for Edge Computing
One of the essential prerequisites of edge computing is low-power consumption. Low-power technology has become an indispensable component of edge computing, affecting everything from industrial production to individual consumption.
Low-power consumption is essential in edge computing because edge devices are installed close to the user or the machine, often in locations where a stable energy supply is not easily available. On the other hand, most of the edge devices, such as smart surveillance and industrial sensors, are required to operate 24/7. Excessive power consumption may lead to frequent charging or battery changes, which could ultimately result in higher operational costs.
To achieve low-power edge computing, Avnet’s DVI system operates on either an NXP i.MX 8M Plus or NXP i.MX9 application processor. These processors have low-power consumption, which meets the requirements of edge AI and machine learning applications. Not only does this edge computing-based, low-power solution effectively improve the response speed and process capabilities of AI models, but it also offers small- and medium-sized enterprises a smart tool with high efficiency and scalability.
From Fantasy to Reality: Welcome to the Future of Smart Manufacturing
Edge AI is redefining the intelligent era of industrial production. From edge computing and limited data training to low-power NPU support, a new era has dawned. Transcending the limitations of mere mechanical operations, the machines can now “think for themselves.” Factories of the future will evolve beyond simple production lines, turning into ecosystems with super-intelligent brains.
A series of innovations and changes have paved the way for a new age of smart manufacturing. Imagine that every machine could “predict” its own failures and repair itself or issue a warning. Production lines would no longer be limited to repetitive tasks but evolve into smart, efficient, automated future factories. Together with Avnet, a growing number of companies are now promoting this transformation, taking steps toward a brilliant future of smart manufacturing.
