Implementing AOI+AI,
The PCB manufacturer reduce the overkill rate and labour inspection costs by 92%
The 21st century is the era of the Internet of Everything(IoT), and electronic products have become necessities of people's lives, thus driving the demand for PCB circuit boards. According to the N.T. Information world ranking, Taiwanese factories account for 5 of the top ten PCB factories, which shows Taiwan's important position in the PCB supply chain.However, in the face of competition around the world, it is an important issue for Taiwanese manufacturers to implement digital transformation tools.
Customer Story
As one of the largest PCB manufacturers in the world, Company A is actively joining the new battlefield of Industry 4.0. To make products 100% flawless, PCB factories often build many inspection stations in the production line, including SPI (Solder Paste Inspection), AOI Before Reflow, and AOI After Reflow. (See the production process below)
The advantage of AOI (Automated Optical Inspection) is that it can replace the manual inspection work in the past. Not only does it never tire, but it can also judge more accurately than the human eye; however, for some elements in the shadow, it is easy to produce false rejects. Therefore, Company A set the parameters to extremely high specifications. As long as the PCB has slight defects, it couldn't pass the AOI inspection.
However, this leads to another problem. Under such high-standard AOI parameters, Company A's Overkill rate reached 50%. That is to say, after the AOI Before Reflow, only 50% of the PCB boards can pass, and the remaining 50% are judged as NG products. Therefore, to avoid eliminating too many qualified PCB boards, Company A has to re-inspect it manually, which has caused a lot of labor and time costs.
Solutions
We analyze the current situation of AOI Before reflow first and set the project goals as:
- Reduce the overkill rate of AOI
- Reduce the workload of manual inspection
Due to the bottleneck of the current AOI accuracy, the Overkill Rate is too high. We decided to reduce the overkill rate by introducing AOI+AI technology. First, each group of New Component needs to collect photos of OK and NG (300 to 500 for each) to build an AI model. Then use traditional algorithms for preliminary analysis, define defects and collect defect data; finally, we collect more defect images through AOI to continuously train our AI model.
Since processing graphics with AI requires powerful computing power, to avoid excessive investment in hardware costs, we use cloud virtual machines (VMs) for model training. The advantage of cloud virtual machines is that they can be flexibly adjusted, and reducing training costs.
Benefits
After implementing AOI+AI, we achieved several goals:
- The overkill rate has been reduced from 50% to 4%
- Reduce the workload of manual inspection by 92%.
- The hybrid cloud architecture allows the cost of retraining new model to be less than 3K USD, replacing tens of millions of server investments.
In the PCB production line, we have been using AOI for ages. However, as PCB become more sophisticated, it is an inevitable trend to implement AOI+AI. AOI and AI complement one another. Data is quickly collected through AOI and used to shorten the training time of AI models. Therefore, AOI+AI has become the mainstream method for many factories to improve accuracy.
Get Ready for Digital Transformation
Based on Cloud, Data and AI technology, WiAdvance provides cloud-based service and vertical SaaS / Solutions
which suit various industries and operational scenarios.
which suit various industries and operational scenarios.