Automating inspection through intelligent systems
01 November 2021
Automating electronics manufacturing inspection through intelligent systems
Automated technologies have never been more important. Manufacturers depend on innovative equipment like machine vision & robotics to keep pace with customer expectations & the increasing speed of business in a digital world.
This article was originally featured in EPDT's H2 2021 Test & Measurement supplement, included in the November 2021 issue of EPDT magazine [read the digital issue]. And sign up to receive your own copy each month.
However, many plants and warehouses still use human workers for visual final assembly quality inspection because often, today’s applications are simply too complex for machine vision, robotics and other automated systems – and not all automation solutions are up to the task of independently verifying products and assemblies for defects. But as Anthony Oh, Technical Applications Manager at electronics manufacturing capital equipment distributor, Altus explains here, there is an answer...
The UK & Ireland electronics sector remains focused on aerospace, military and high-value electronics, largely because of our traditional strengths in quality and security of IP. We have large swathes of products that are ‘safety critical’, where it would be unacceptable to face field failures. This goes for PCBA assemblies, but also final assembly of housings and the box build process.
Traditionally, this has had to be completed by highly skilled quality engineers, and is time-consuming and subject to human errors that are unavoidable across a repetitive function. Like SMT inspection, which has been automated through solder paste inspection (SPI) and automatic optical inspection (AOI), it’s now time to remove the next – and more complicated – cost driver of manual final assembly inspection.
There is an answer, however. A new hybrid approach that combines 2D, 3D and deep learning technologies is quickly, accurately and reliably enabling industrial applications once considered too multifaceted or difficult for automated inspection – and delivering huge cost savings in doing so.
AI for intelligent electronics inspection systems
The price to pay for poor quality
Humans make mistakes in manual visual inspection: for example, tired workers might miss defects that escape quality screens on the production floor. When defective products make it beyond the factory, manufacturers pay a steep price. The ‘cost of poor quality’ includes returned or rejected goods, scrap, rework and repair, and, in many cases, the negative impact on brand reputation and customer satisfaction. In many supplier/customer relationships, a single defective product could lead to rejection of an entire shipment and potential financial penalties.
Seeking to get ahead of such issues, many manufacturers have turned to smart cameras, artificial intelligence (AI), robotics and big data analytics as part of an Industry 4.0 and, more specifically, Industrial Internet of Things (IIoT) strategy. Automated visual inspection systems produce data useful for identifying the causes of production defects, leading to efficiency improvements and corrections. A comprehensive database populated with inspection results, including information on defective and defect-free products, provides manufacturers with actionable intelligence. With the help of big data analytics technology, manufacturers can analyse and track quality, trends, common defects and evolving quality issues, as well as proactively introduce improvements in the production process, product design and supply chain management.
Flexible inspection technology
While many original equipment manufacturers (OEMs) adopt machine vision and automation systems to keep up with customer demands, certain situations move beyond the scope of the traditional robot, machine vision and motion-control technologies that have proven successful in the past. For example, many manufacturers have adopted mass customisation and personalisation into production processes. Doing so forces these companies to account for increased production environment variance, which can throw a wrench into traditional machine vision processes. Quality inspection routines for such variations may include multiple algorithms applied in a specific sequence to extract useful information from images. Traditional rules-based algorithms define defects using mathematical values and logic rules, but using such algorithms to create an accurate and reliable inspection routine — free of extensive false negatives and false positives — can take hours or days, depending on the product and the programmer’s skill level. Multiply that time requirement by 100 or 200 product variants and it’s easy to grasp the scale of the challenge.
Furthermore, defining complex assemblies and shapes using mathematic values results in a rigid rule set that may not offer the best solution for modern product lines. Manufacturers require not just automated inspection systems, but also flexible inspection technology that adapts as products, processes and environmental conditions change.
Intelligent PCB inspection [shutterstock: 600056555]
Analysing with deep learning
With deep learning, an expert trains software using images of ‘good’ and ‘bad’ parts. In manufacturing, the expert may be an engineer or a machine operator well versed in the operation of, and potential defects generated by, production equipment. Deep learning software statistically analyses the images for features and relationships between features. It then creates a weighted table or neural network that defines what makes a good or bad part.
While it sounds simple, the training process for an automated assembly line inspection solution involves intensive computational analysis. During inference, the system uses considerably less computational power to apply what it learned during the training phase. Lower computational needs during inference are good for OEMs that require in-line inspection, but that cannot leverage remote server farms, due to delays in data communication. Instead, these OEMs rely on local PCs and embedded computational systems to meet inspection speed and production throughput requirements.
To simplify the deep learning training process, AI solution providers typically supply partially pre-trained neural networks designed for specific purposes, such as optical character recognition (OCR), reading damaged or distorted barcodes, or even automatically inspecting medical X-rays. Deep learning software suppliers offer additional neural networks for 3D parts, enabling the software to identify random scratches on smartphone cases, for example, or missing components on a printed circuit board, even where every component might look surprisingly different depending on insertion angle, location on the board or lighting effects.
One company that has a solution to this training process is KITOV.ai. Their flagship standalone deep learning product inspection station comes standard with several pretrained neural networks for locating and inspecting screws, surfaces, labels, optical character recognition (OCR) and data ports.
The KITOV ONE platform is not designed to replace AOI, but it has been specifically designed to replace human operators and reposition their skills to more value-added processes. Creating a flexible final assembly inspection cell has been the challenge, especially when considering the high mix, low volume region the UK & Ireland is – but we feel the KITOV, mounted to a 6-axis moving robot and simply programmed from current inspection protocols provides an answer. From what we have seen, the return is clear for those interested to take the leap.
KITOV.ai intelligent inspection station
Faster inspection, fewer escapes (devices that are passed in the testing phase, but are, in fact, defective), more consistency, less skills required and automation of process are all benefits of the platform.
The hybrid approach
Inspecting highly variable, multi-component finished products and assemblies, such as electronic equipment, medical kits and automotive parts, presents a significant machine vision challenge. Traditional, rules-based systems produce unacceptably high false-negative and false-positive rates. Using only deep learning to solve the problem involves training the system to recognise every component in an assembly, and then to combine the components into a single assembly for the final quality inspection step. In this case, achieving acceptable false-negative rates without allowing too many defective products to escape the quality check often proves difficult, even for experienced vision system designers.
KITOV ONE combines the best capabilities of traditional machine vision algorithms with deep learning capabilities to enable the inspection of complex assemblies, and to continually adapt to changing conditions. It essentially allows the system to learn what makes good and bad parts during production runs, and not just during the training phase of system development, placing automated continuous improvement of industrial processes within reach.
We expect to see more of the units in the UK after KITOV made an excellent start with many installations in Europe across 2020, and as process engineers’ focus returns to capital expenditure for high return processes in the pandemic recovery. We have already seen our first project where assembled product AOI is part of the scope, and expect to see more similar demands as high-quality sensitive OEMs realise that flexible processes are available to reduce their risk.
Contact Details and Archive...