AI Vision in Automated Optical Inspection

Back To AI Computing Platform   06 September 2564

An important step in any kind of manufacturing is inspection and defect detection. In manufacturing of printed circuit boards (PCB) the task is carried out with machine vision controllers in and application known as Automated Optical Inspection, or AOI for short. However, with increasing complexity and use of compact components and boards, the task is becoming more difficult for machine vision systems to keep up. Machine vision is often thought of as a kind of AI, when in fact machine vision is not actually intelligent. Machine vision requires every step and condition be programmed into the system, what is known as hard coded intelligence. As PCBs become more complex, the program can become more complicated and bulky, resulting in errors. With visual inspection, it becomes impractical and almost impossible to try to program every way in which a PCB could have a defect or fault. One company is creating an AI solution to powering AOI. Their solution offers reduced time to setup and deploy, as AI can be trained to identify faults and defects. AAEON has been working closely to help build and develop the hardware to power their AOI application AI.

Challenges

AOI is an intensive and demanding application. It requires not just high performance software, but a system that is equally powerful and high-performing. This creates several challenges to developing the system.

High Processing Demand

In order to process images and signals quickly and accurately, AOI requires a system with high computing power and processor performance.

Visual and AI Processing

To fully meet the needs of AOI, the system needs to support expansion cards, such as graphics/GPU cards or neural accelerator modules to process images and power the AI system.

Cameras and Sensors

Different machines have different configurations and components, from the numbers of cameras and sensors to the type. To meet the full potential of an AI powered AOI solution, the system needs to support a range of I/O connections.

Solution

AAEON worked closely with the developer and through our manufacturer services provided the best solution for the task. The BOXER-6841M with AI Core XP8 was chosen as the best solution for AOI powered by AI.

Intel Processors

The BOXER-6841M supports 6th and 7th Generation Intel Core processors, as well as Intel Xeon server grade processors. Paired with up to 32 GB of memory, the BOXER-6841M offers high performance processing needed for the most demanding applications.

Expandability

The BOXER-6841M offers a range of expansion bay configurations, including PCIe [x16], PCIe [x8], and PCIe [x1]. This provides users the ability to add in expansion cards to help power Edge computing applications, such as the AI Core XP8 module featuring eight Intel Movidius Myriad X VPUs.

Impressive I/O

The BOXER-6841M is designed for embedded industrial applications, and comes with an impressive range of I/O ports to easily integrate into any project. With five LAN ports and four USB 3.0 ports, it can connect to a wide range of cameras and sensors for use in AOI and other applications.

Impact

With the powerful BOXER-6841M, the developer could push their application to its full potential, creating an AOI system that can inspect multiple product lines without need of reconfiguring, and can be quickly trained for new products or to identify previously unknown defects.

The system will help PCB manufacturers save money and increase productivity through greater accuracy and fewer mistakes and false positives than traditional machine vision based AOI, and also happier customers for those manufacturers.

Product

The BOXER-6841M system from AAEON is a powerful embedded industrial PC built for the most demanding of machine vision and AI Edge applications. Supporting 7th generation Intel Core and Intel Xeon Server Grade processors, the BOXER-6841M features up to 32GB of memory. The BOXER-6841M range offers a combination of expansion options, including PCIe [x16], PCIe [x8], and PCIe [x1]. This allows for expandability with frame grabber cards for machine vision applications, or AI acceleration with cards such as the NVIDIA GeForce GTX 1080 or the AI Core XP8 from AAEON.

The AI Core XP8 is a PCIe [x4] expansion card featuring eight Intel Movidius Myriad X modules. This innovative module offers speeds up to 840 fps (560 typical) to help power the most demanding of AI applications. By offloading AI inference onto the AI Core XP8, it helps free up the system processor to handle other simultaneous tasks, providing an increase to overall system performance. The AI Core XP8 is compatible with Intel OpenVINO toolkit and AI frameworks such as TensorFlow.