Development of an Imaging AOI AI Ecosystem

Call-for-Participation Webinar

Statement of Work (August 3, 2022)
Project Statement (August 3, 2022)

Current industry practice: every EMS company has its own AOI image data, which is private and isolated
Electronics manufacturers are facing the challenge of insufficient image libraries (data sets) to train AI models for AOI and AXI, which can result in inaccurate AI models for image classification purposes
Image data collection — data acquisition, identification, labeling, etc. — is typically time consuming
There is no collaboration model for sharing image data among the EMS players, balancing the benefit and risk mitigation

Create an industrial level collaboration to construct/share image resources (in an appropriate way) to develop AI models

Proposed Strategy/Approach: Establish an Ecosystem to Collect and Share Image Data
Individual companies will contribute their respective image data sets to a central data repository where data will be consolidated into different categories/types to input to the AI training cycle, significantly increasing the scale of the image data
With the consolidated data sets in the repository center/server, these data could be further trained into different AI models for future service requests from companies
Companies can also request either a data set from the repository for its own AI model training purpose, or classification service from the data/AI repository server via RESTful API request
With this shared data/AI classification service, the accuracy of the AI model could be significantly improved, and the AOI process performance could be further improved with this new type of AI model

Mark Schaffer, INEMI
[email protected]

Wayne Zhange, IBM