Even though inspection is considered as a non-value added activity, our processes are not stable enough to eliminate inspection completely. Even when we all are working continuously to improve the process stability and capability, eliminating inspection is still very far from reality.
Manual assembly, even at its best, is still vulnerable to errors because humans inherently produce variability. It’s just what we do. To negate this variability, quality control methods typically depend on inspection programs. Whether programs are designed at the same time as the products or after the products are in production, the inspection process is developed with four critical things in mind:
- What needs to be measured with qualitative or quantitative methods (size, color, placement of components, pattern, etc.)
- What pass or fail criteria will be used for an inspection (grey discoloration, width of 1.2 centimeters or more, etc.)
- How inspections should be conducted (random sampling, results that need to be collected, what happens to products when they don’t pass, etc.)
- Where inspection results go to provide feedback on manual assembly processes (revise SOPs, training updates, QMS documentation, etc.)
Automated vision inspection systems aren’t developed for assembly feedback
Vision inspection systems are developed for the product and the inspection criteria they evaluate and the output results in only one direction: downstream. For example, a vision system that can detect missing components needs to work from a known, well understood defect catalog specific to the product. It’s programmed to detect that product’s components and their locations.
As a highly specialized system used for specific detection, it will only output binary details – either components are missing or they’re not. Knowing a component is missing is extremely valuable in downstream production, because it can still be fixed.
It’s a perfect example of how inspection does its job of quality control well. It’s also a good demonstration of how the inspection system’s output – on its own – can’t prevent defects from happening or meaningfully reduce scrap rates.
While this is a limitation of any inspection system, specialized vision systems are getting smarter with new technologies. AI and deep learning techniques are providing solutions that make automated visual inspection work in more complex scenarios. Have a couple of versions of the same product? With robust computer vision solutions, that’s no problem.
Advanced vision inspection systems that incorporate learning algorithms move past limits of traditional systems. They don’t solely rely on a catalog of defects. They learn from models that can provide more meaningful information – textures, deformations, cosmetic issues, verification of part placement and more. And they’ll learn from sampling thousands of images to detect granular details. These types of anomalies can then be better classified and more details are sent to people and systems downstream to correct problems before they escape. Yet, if feedback is needed upstream to the people and processes that created the issue, you’re out of luck.
Vision systems unite: AI technology to rescue your inspection program
This is where AI technology, used in manual assembly, is helping to holistically round out today’s inspection programs. Video traceability systems such as Drishti can close the gap on defect detection and show why a defect has occurred. It provides a video system of record for all manual processes used to build a product. And Drishti’s AI models turn that video into data and analytics to measure cycle times and count steps in standardized work processes.
You want to understand why you’re seeing a particular reject rate, or know more about a new undetected flaw that caused a line stoppage? If assembly is suspected, simply look up the video by serial number or date and timestamp, and you’ll have the replay of the process at your fingertips.
It compliments your vision inspection system, both traditional and advanced. And it completes the feedback required in all inspection programs, leaving no open loops or loose ends. Complementing traditional inspection systems with AI technology provides a comprehensive check on the units leaving your assembly line, ensuring the highest quality.
Sandeep Nagaraju is the head of remote industrial engineering services and program management at Drishti. He has more than 12 years of experience in implementing industrial engineering and lean manufacturing tools in the factory. Sandeep is also a certified coach and assessor for lean manufacturing. At Drishti, he is helping customers to identify the potential improvement opportunities in their manufacturing lines using Drishti. Sandeep is based in Bangalore, India.