Printed Circuit Boards (PCB) have played a significant role in transforming electronics. PCBs provide a platform where miniature electronic components are displayed and interact with one another. However, the requirements of these boards are becoming much more stringent electronic components develop and offer great capabilities. It is vital, for example, to avoid any electrical interference and ensure electromagnetic compatibility.
To this end, The Fraunhofer Institute for Applied Information Technology (FIT) has developed a solution that caters to all PCB requirements — and with higher efficiency. It is a modular AI platform that optimizes the design and testing of PCBs, shrinking the outlay by about 20 percent.
Optimum space is mandatory in PCB designs. In fact, an engineer needs to position all components as close as possible without risking its working efficiency. This process relies largely on the experience of engineers, whose designs must then be tested in real trials. A further complication is that results are not stringently documented — meaning that error-prone designs undergo repeat testing, which leads to increased costs.
PCBs need to be manufactured as per exact specifications and any small error can easily lead to rejection by the Automated Optical Inspection (AOI). The AOI uses various techniques to analyze the accuracy of manufactured PCBs and to ensure it’s void of defects.
This method often results in a high false-negative rate which implies that many fully-functional PCBs are getting classified as defective. But if this rate is too low, it means that defective components are entering the supply chain. Under both conditions, the follow-up rates increase as all pieces then require re-inspection and testing. It’s not easy to get balanced, positive or false negative rates based on human inspection since human errors also have a part to play in this process.
The modular board developed by FIT provides a glimpse of the future of inspection processes. The quality of algorithms’ decisions is here improved with camera recorded images of the PCB. The deep learning and machine learning modules use previously fed data to analyze the quality of PCB.
“The modular design means we can harness several algorithms, which continually enhance their own performance,” explains the project manager of Fraunhofer FIT, Timo Brune. “Data generated by ongoing automated inspection of components flows back to the algorithm. This then provides the basis for a process of self-learning by the artificial intelligence module.”
According to Brune, this permanent feedback enhances the database and optimizes the true negative rate.
“Early estimates from industry indicate that this could reduce the use of production resources by around 20 percent,” he adds.
These modules can be trained by users with their own process and production data. These algorithms can also be used for designing new PCBs post-training. This reduces the time to market and higher costs due to trial and error methods.