Driving quality inspection automation in a post pandemic world
Quality drives success, and this holds true for PCBA manufacturers - large companies producing for their own products, as well as contract manufacturers working for a variety of customers. While until now high quality assurance has been the province of human skilled supervision, constraints on workforce availability and the need to avoid social proximity on the production line are imposing new challenges in an already very competitive industry. The answer to these new challenges: AI and automation.
By Massimiliano Versace, Ph.D., Co-founder and CEO, Neurala
Vulnerabilities coming to light
For PCBA quality inspection, the status quo has typically consisted of x-ray inspection, functional testing, and AOI machines (Automated Optical Inspection). Factors like product specifications, or individual customer contracts in the case of contract manufacturers, determine the level of testing. The more testing that is done, the more expensive the PCBA, and the higher the quality. But AOI machines have limitations - they need to be specially programmed, requiring a trained operator to function. And when changes occur, reprogramming can take considerable time, causing delays in production.
In addition to the operation of the machines themselves, PCBA manufacturers are struggling with workforce challenges. Traditionally, quality inspection by AOI machines and functional testers is a very labour-intensive process, requiring a person to put the PCBAs on the bed of the AOI machine to be analysed. From there, someone else reads the outputs and determines product viability after receiving results. One alternative might be to use conveyors to move the product while a controller decides what to do with the product after inspection, but AOI machines are not compatible with this level of automation.
The National Association of Manufacturers is reporting that 53.1% of manufacturers anticipate a change in operations due to the health crisis.
Reduced workforce, the need to speed up reprogramming, and new constraints on workers proximity are all driving forces towards adoption of automation and AI able to provide human-level precision in quality inspection and the flexibility that today’s solutions lack.
Complementary, enabling technology
But what is the price tag associated with these needed innovations? Changes to the production process, shifts in required skillsets, and capital equipment expenses often limit the reach of disruptive innovations, which result in slow and scattered adoption.
This will not be the case with AI and deep learning. While these technologies have been talked about for years, only recently have practical, real world applications for manufacturing been dominating the headlines amidst the rush to respond to the pandemic. Vision AI software is one of those applications, improving production efficiencies and quality control with fewer people on the factory floor.
The integration of AI and deep learning in existing processes can be simple and inexpensive. For example, if AOI machines are currently in use and generating images, a good vision AI system will not require a new system to be purchased and installed. In this case, instead of relying on the user interface to set up a GigE camera, APIs can be used to pull images in for analysis. Ideally, the only necessary new piece of hardware is an inexpensive computer for the image analysis to occur.
Flexible vision AI systems are a great option for scenarios where creating models for the AOI machine is too time consuming, or when an existing model was not performing to a high enough degree. This enables manufacturers to improve performance while decreasing cost when small batches of PCBAs are being manufactured. Because it’s important to keep changes on the floor minimal, an operator still manages the inspection station, and the Modbus outputs are used to create a message to an HMI that tells the operator if each PCBA is acceptable or defective.
If a manufacturer is looking to avoid AOI machines - for example, a company that is just starting to manufacture PCBAs where an AOI machine is too large of an investment - the right vision AI solution could offer a significantly smaller investment for quality inspection. Here, a GigE camera with appropriate lighting would be purchased and set up at an inspection station. In order to have good enough resolution, a mid-range GigE camera is typically necessary.
A fixture should be used to ensure reasonable repeatability in placement of the PCBA. An Industrial PC should be all that’s needed to run the vision AI solution, and communicate using OPC UA (the next protocol available after Modbus TCP). At the beginning, the outputs are used to communicate with an operator manning the inspection station.
The advantages of implementing vision AI for PCBA manufacturers are clear:
Ease of implementation: a software solution can be integrated to fit the needs of the manufacturer, whether it’s working with installed AOI machines or implementing cameras at an inspections station.
Flexibility: With vision AI there are a variety of ways to implement quality inspections - from large volume to small batches - and manufacturers decide how they want to integrate the software.
Very little data needed: In the past, AI models needed thousands of images covering all expected defective variables (a difficult task). Emerging advances in vision AI only require images of 'acceptable' product to train a vision AI model to find defects.
Rapid ROI: The increased rate of set up, and the fact that skilled programmers aren’t needed, decreases overall cost of implementation of an AI system and accelerates the ROI.
Over the years, printed circuit boards have fueled major electronics revolutions. It is only natural that PCBA manufacturers will be among the early adopters in a new paradigm-changing transformation which will see AI and deep learning provide previously unreachable production quality and flexibility.
As with so many innovations in history, this shift is born out of necessity. Welcome to the ‘new normal’ in PCBA manufacturing.