Building a Powerful Visual Quality Inspection Solution Using a 5G Network
By: Dave Dolson
How Quantitative Explainable AI can make human visual inspections more productive
Increasingly, manufacturers are implementing AI into their production process for cost savings and improved product quality. To make these AI systems more efficient, resilient, and manageable, it becomes essential to network them together as a distributed system. The availability of 5G makes high-speed low-latency networking easier to deploy than ever before because there is no need to run cabling. The intersection of network-enabled AI and 5G is a powerful combination.
An AI inspection solution for manufacturing requires a combination of these components:
- Cameras and lighting, mounted on the line.
- Control and capture software.
- AI inference engine(s) loaded with trained AI model(s), sometimes using GPU hardware.
- Disk storage of part images for traceability and analysis.
- A database for collection of results.
- Various user interfaces for stats & monitoring, providing feedback, and configuring the system.
Although it is completely feasible to deploy all of these components as part of each assembly line, the costs of doing so may be higher than is strictly necessary, and managing multiple lines can be chaotic. In contrast, having a robust network gives you the freedom to deploy these components where they are safest and most cost-effective.
Here are some scenarios where you can improve costs, manageability, and data security:
- Use network storage (NAS) or cloud storage for highly reliable image and data storage, avoiding the precarious state of files stored on local machines.
- Use a single database in a server room to collect data from all lines, protecting against data loss on the factory floor and improving manageability.
- Use inexpensive cameras and commodity industrial computers “at the edge” (on the assembly line), sending the images over the network for AI processing away from the hazards of the factory floor.
- Share AI computing engines across multiple lines, for cost savings and better maintainability.
- Using Internet services techniques, share a set of redundant AI computing engines with multiple cameras and lines, adding resilience against individual hardware failures, and allowing rolling upgrades without stopping production.
- Give different users different interfaces: let the operator provide task-specific feedback from a touch-screen on the line, while keeping the data in a server room, and letting managers review the statistics and results from all lines from the comfort of their offices.
Having network access to all lines provides additional levels of control:
- The ability to have known and consistent configuration and software versioning across your production lines.
- As AI models learn, ensure the best models are deployed consistently across production lines.
- Orchestrating highly-customized, small production batches across multiple lines.
It can be tricky to cable Ethernet into factory environments, especially if the equipment is often moved around. 5G promises to bring the speeds of wired networking to any devices reachable by radio waves, with 3 frequency bands addressing various trade-offs, leaving no excuses for having equipment off the network.
Of course any networked solution should be built on a solid security framework. Look to encryption technology, again from web development best practices, to create very secure systems.
DarwinAI manufacturing solutions are built using the architecture of web applications, on a Docker-based software stack that is ready for network deployment now. We deploy computing resources at the most appropriate physical locations, connected by the speed and reliability of 5G.
If you want to know more about our visual quality inspection system using a 5G network, contact us.
Image Credit: Ivan Radic