How does AI contribute to continuous improvement?

Depending on whom you talk to, artificial intelligence (AI) is likely to open amazing possibilities or lead to the end of humanity as we know it. The reality is that when appropriately used, AI eliminates a lot of repetitive tasks that have high levels of variation and cost when done manually. In the quality realm, this opens the door to a discussion about whether a truly automated inspection process is a non-value-added or necessary non-value-added activity.

Manual inspection is costly, and accuracy can vary widely among operators, the time of day the activity is performed, or even the day of the week the activity is performed. Quality philosophy has long held that it is better to prevent defect opportunities (quality assurance) than to try to inspect them out (quality control). If a printed circuit board assembly (PCBA) can be 100% visually inspected by machines during SMT and secondary assembly processes without significant throughput time, however, does it make sense to do so? Let’s look at the pros and cons:

Potential benefits include:

• Faster identification of defects earlier in the process, which translates to less rework activity to address the issue
• Inspects against a standard that improves as the machine learns from past experiences
• A check and balance against defects unlikely to be caught in test, which pays for itself when considered against the cost of field failures and customer returns
• Inspection of a PCBA that takes 2-3 min. manually is less than a minute by machine.

Potential disadvantages include:

• Requires machine time and engineering resources
• Won’t detect component failures or identify issues in hidden solder joints, so additional testing remains necessary to ensure no defects escape the factory.

At lower volumes, the 100% inspection strategy is unlikely to justify costs. At higher volumes, however, the costs of even a small percentage of returns improves the cost equation. AI tilts this benefit even further by increasing the inspection complexity that a machine can perform. This enables inspection beyond components and solder joints.

In 2022, SigmaTron International’s facility in Tijuana, Mexico, began exploring the best way to automate inspection and integrate the captured data into real-time corrective action throughout its entire PCBA assembly process. A July 2022 PCD&F/CIRCUITS ASSEMBLY column, “An Industry 4.0 Approach to Employing 3-D AOI on an SMT Line,” discussed the journey of integrating Industry 4.0 capabilities in a Lean Six Sigma framework in this facility’s SMT area. A September 2023 PCD&F/CIRCUITS ASSEMBLY column, “Continuous Improvement and Mass Inspection,” looked at the challenges of replicating automated inspection in secondary assembly operations.

The initial goal in utilizing 3-D AOI in secondary assembly operations wasn’t to do 100% mass inspection. Instead, it was to focus on the projects with the lowest yields to drive a continuous improvement effort to reduce defect opportunities. As the machine’s manufacturer upgraded AI module software capabilities, however, it became apparent that more complex inspections were possible, and implementing a 100% automated inspection process was less costly than having manual inspections on areas of the PCBA not traditionally inspected by AOI.

While the end goal is to expand the 3-D AOI’s ability to inspect 100% of the board, new inspection steps are being added based on yield trends and customer feedback. For example, labeling issues prioritized the creation of an algorithm designed to verify that not only were labels correctly placed, but also that the correct characters were in place on each label. The normal process for label inspection is just to read and keep information in the database. There were no parameters set to check for the correct information and characters.

Working with the machine’s manufacturer, the engineering team created an algorithm that checks keywords such as the correct part numbers and the correct quantity of characters. If the program detects any inconsistencies, the test stops and sends a message.

The team utilizes the C# programming platform to create each new algorithm. Once the machine begins inspecting to the new parameters, production operators or the engineering team can fine-tune based on yield trending. The machine learns from corrections as false calls are identified. Over time, it analyzes its library to determine whether images fit the parameters and will query the operator on situations where clarity is needed. Once the number of false calls drops significantly, it will query to see if the improved parameters can be transferred to other machine libraries, making propagation across equipment easy.

Inspecting in secondary assembly is still more challenging than in SMT. For example, components added in secondary assembly may be too tall for the vision system. The shape of some PCBAs can also create vision system issues. In these cases, changing camera placement or programming can solve the issue.

Broadening the scope of inspection to include elements of the PCBA not traditionally checked by automated mass inspection equipment doesn’t actually add an inspection step. Instead, it increases the benefits of the inspection step in terms of the number of items inspected. At the same time, broadening the inspection parameters also broadens the available yield trending data, improving continuous improvement teams’ abilities to quickly identify areas of needed process improvement and initiate corrective action.

Filemon Sagrero is continuous improvement engineer at SigmaTron International (sigmatronintl.com) and a Six Sigma Black Belt; filemon.sagrero@sigmatronintl.com.

• Next