Inspection Bottlenecks in PCB Assembly

The supply chain for electronic manufacturers is fraught with deep vulnerabilities. This is primarily due to pandemic-related disruptions and geopolitical tensions. According to Rush PCB Inc., the global shortage resulting from these vulnerabilities is also affecting governments and businesses that rely on these manufacturers.

This global shortage is impacting the production of everything, including critical infrastructure, medical devices, and even automobiles, mainly due to the shortage of semiconductors, essential electronic components, and printed circuit boards. In turn, manufacturers are exploring various ways to increase the efficiency of the production processes of such components as a means of mitigating the risk of shortages. The government is enacting legislation like the CHIPS Act to boost domestic research in and manufacturing of semiconductors.

Although the CHIPS Act is primarily about semiconductors, it applies equally to PCBs while complementing international industry standards like IPC-1791. It primarily combats poor quality of components and their counterfeits that primarily reduce the throughput of PCB assembly production.

Inspection Bottlenecks in PCB Assembly

Manufacturing PCBs is a complex process involving both humans and technology. Everything, ranging from aircraft to microwaves and toasters, requires these ubiquitous boards to perform their intended functions. Depending on the application, PCBs can be highly varied—they can have different sizes with complicated geometries and diverse components—necessitating specialized skills for their inspection, which, in turn, can be a laborious and intense task. For instance, although a conformal coating can protect a PCB from environmental factors, it can also mask several defects that a human can find painstakingly difficult to identify.

Even when PCB inspectors are proficient, manual inspection of PCBs can lead to high COPQ or cost of poor quality. This is mainly due to the additional expense of uncovering failures in later parts of production and reworking them, often leading to high scrap. Furthermore, as highly experienced inspectors retire, there will be fewer qualified personnel available to replace them.

It is practically impossible to close this gap in the short to medium term. This is despite adequate funding, good intentions, tax rebates, or recruitment videos. Although the CHIPS Act envisages ambitious production increases, this is not likely to be fruitful soon, as newer facilities will take years to produce meaningful outputs.

Manufacturers are, therefore, turning to transformational and disruptive technologies to mitigate these challenges.

VQI As a Replacement For AOI

So far, manufacturers have been primarily equipping assembly lines with AOI or automated optical inspection systems. These systems relied primarily on visual comparison of an assembled board to detect differences in comparison with a known good board and a computer to highlight these differences to the operator.

With the advent of AI or artificial intelligence, VQI or visual quality inspection systems are replacing the older generation of AOI. While VQI does keep humans in the loop, it is a highly automated system and easily outperforms AOI processes.

How VQI Works

A primary requirement for VQI to be effective is a golden board, certified by a human operator, to be the reference or ground-level truth for all types of inspection. In comparison with the long-term programming necessary for AOI systems, this step hardly takes a few minutes.

Each production PCB goes past one or more cameras that capture multiple images of the board.

A computer compares the images with that of the golden PCB. Although this step is similar to the AOI system, the difference lies in the use of an AI engine to identify the variations and classify them as defects. In comparison with the performance of a human operator, the AI engine works much faster, taking only a few seconds.

The computer uses an intuitive user interface to present the AI findings to the operator. It can also highlight the reason for the AI’s classification of a specific anomaly as a defect.

The operator can either validate the decision of the AI or override it. Each decision validates the AI to perform at higher levels of accuracy for future inspections.

VQI results in a highly accurate, more automated, and non-destructive inspection process. In conjunction with a few human experts, it can give much better results, improving efficiency substantially while reducing the cognitive burden on humans. Furthermore, it is possible to archive the image and the accompanying data for each PCB for more powerful analytics.

Analytics for Lowering Data Barriers

Implementing modern AI systems requires overcoming a few key obstacles. These typically range from dependence on training data to the effort and overheads of collating large, labeled databases mostly required by the machine learning systems. Many manufacturing contexts do not make this easier, mostly because it is difficult to obtain relevant images of component defects.

Manufacturers are resorting to proprietary techniques for training VQI systems with substantially reduced amounts of data. Thereafter, the system enters a feedback loop, which continuously refines its capabilities as it trains the system. Very soon, the system approaches the accuracy of a highly skilled operator, subsequently surpassing human limits of expertise.

Apart from making the AI smarter, the feedback also provides the manufacturer with adequate amounts of data, which they can subsequently feed into analytics tools. They can use the information to generate root cause analysis for implementing design and process optimizations.

VQI Strategy Implementation

VQI systems offer straightforward benefits. However, given that AI in manufacturing is a relative newcomer, deploying it in production may not be an easy task. Organizations may have to consider adopting VQI to transition to this powerful technology.

To start with, technology leaders may have to demonstrate quick wins with VQI to obtain organizational buy-ins. For this, they can execute pilot inspections and generate concrete ROI to demonstrate the economic benefits of VQI solutions before implementing them more expansively.

Next, they may have to deal with enterprise requirements like governance, security, and IT when implementing VQI systems into production, as there may be special considerations for unique aspects of the solutions from AI systems. Machine learning systems typically have non-deterministic aspects that may require updated infrastructure technologies concerning stress testing and versioning.

Finally, technology leaders may also have to consider whether they will build the system themselves or buy it. There are pros and cons for both methods that will require careful weighing. If the manufacturer has adequate expertise and resources, they can decide to source the AI engine and build the rest around it themselves. Others may find it easier to buy an off-the-shelf solution that they can simply fit into an existing production line.

While some manufacturers may want to start right away with VQI implementation, others may foster in-house teams with a view to long-term advantages over competition.

Conclusion

Although it has taken a long time to arrive, AI is finally transforming production system workflow in tangible terms while delivering meaningful ROI. According to Rush PCB Inc., not only can Automated VQI help manufacturers mitigate several risks in the short term, but it also enables them to take advantage of other transformational technologies in the future.