Automatic X-ray Inspection (AXI) is widely used in the PCB assembly process to detect manufacturing defects as well as provide feedback for process improvement. With the proliferation of BGA and PoP packages, the importance of AXI is gaining more prominent. However, maintaining a robust AXI process is a challenge because it relies on human operator to make the final judgment to either Pass or Fail a board. It is a norm for human operator to review few hundreds of false calls images on each board. Base on observation, operator only spend about 1s to review each image so as to keep up with the output which could lead to escapes. Escapes could also be due to human fatigue. Not only that, human judgment varies depending on the knowledge and experience.
Lately Artificial Intelligence (AI) is in vogue. Various solutions have been thought of revolving around AI. Although AI is an emerging technology, there has been tremendous breakthrough in hardware, software model coupled with huge amount of dataset generated by factory, we think that AI can be an effective tool in improving the factory PCB assembly process.
We initiated this project to demonstrate that the number of AXI false calls to be reviewed by operator can be reduced with the introduction of AI. Our AI model is able to hit above 90% accuracy. It is also able to detect some of the escapes judged by the operator. We will share what are the strategies and preparation needed to implement AI. We will discuss about the quality of dataset and its impact to the accuracy of the AI model. Also shared are benchmark results of the various AI models.
We would also like to propose to the industry on how we can accelerate the adoption of AI to improve the factory PCB assembly process.