AI Case Study
Jabil eliminates false-positives with automated optical inspection of its manufacturing processes using deep neural networks
Jabil has collaborated with Microsoft to leverage Azure Machine Learning Workbench. The solution extracts real time information from the assembly line to optimise the manufacturing and and quality control processes, using a predictive model. It enables Jabil to achieve manufacturing with more speed and to anticipate issues or proactively deal with them. During early testing, the solution predicted non-defective components with 75% accuracy and successfully identified 92% of defective components.
"Project Brainwave is a system that uses field programmable gate array (FGPA) to efficiently do calculations rapidly (with little lag time) and economically, with the potential to improve the image processing rate to 550 images per second.
By using FGPAs, Jabil hopes to refine its AOI process, using the data the system collects to hone its ability to spot defects and ensure that only truly defective components require human operator inspection. Leveraging AI would allow operators to focus on more value-added tasks that machines cannot complete, and the feedback from operators would also be fed back into the system to allow it to learn and further perfect its accuracy.
Another attractive aspect of the solution is that it is easily scalable; one AI engine can collect data from multiple operator stations, so there is no need to have one engine per station. The investment is also relatively small, its main components being edge computing hardware and data storage to allow the system to train and retrain itself."
"In Jabil’s early tests, this solution accurately predicts non-defective components with 75% accuracy and correctly identified 92% of defective components."
"The Project Brainwave preview includes the ability for customers to do ultra-fast image recognition for applications such as the one Jabil is piloting, and it lets people do AI-based computations in real time, instead of batching it into smaller groups of separate computations. It works on TensorFlow, one of the most commonly used frameworks for doing AI calculations using deep neural networks, a method that is roughly modeled on theories about how the brain works. In addition, Microsoft is working on building the capability to support Microsoft Cognitive Toolkit, another popular framework for deep learning." (Microsoft)
"When Jabil manufactures certain components, it uses a traditional computer-based system to perform AOI of those components to ensure quality.
Those AOI systems require a skilled engineer to build and hard-code algorithms to help the system identify a good component versus a defective one. These systems, however, do not have the ability to learn or adapt, thereby resulting in a high percentage of false positives – situations where a component is flagged as defective but is not.
When a component is flagged as defective, an operator is required to manually inspect it to confirm the diagnosis. In a situation where one station produces 2,000 components daily and the AOI has a false positive rate of 30%, it can result in operators spending up to 200 minutes daily doing manual inspections.
Recent competitions, however, have shown that deep neural networks are more accurate than humans at image classification tasks and can process them exponentially faster (40 images per second)."