Industry Insights

Make CSI Part of Your Manufacturing – IBM Visual Inspection for Quality

Okay, I admit it. I am a huge fan of CSI (Crime Scene Investigation, originally ran on CBS from 2000 to 2015, spanning fifteen seasons) TV series. I hardly missed any single episode of the shows regardless the crime was committed in New York, Miami, or Las Vegas. What really attracts me is not the cast of actors or actresses in the episodes but the director is able to present to the audience a scientific, logical, and systematic process to solve the crime through small traces such as a sheer drop of blood, a fragmented finger print, or an unnoticed fiber drop on the floor found in the crime scene all the way to find out eventually who really is behind the smoking gun! Well, if you ask me what does not make sense in any of the crime stories, I would say solving a crime of any kind in just 45 minutes in time is the only one (to be honest, I would love to see a two-hour episode but I guess CBS just never gave that much time to the show).
It does not really matter whether you are so crazy about the TV series like I am or not, CSI and its crime labs always play a vital part in modern crime investigation and solving. It might sound insane but have you ever thought about acquiring a highly effective CSI team for your manufacturing shops to find out who or what commits the crime of making the bad (defected) products that eats up the unnecessary manufacturing cost and your clients’ satisfaction? Traditionally, a combination of Automatic Optical Inspection (AOI) and human visual inspection approach is deployed for product surface inspection as part of the QA practice in most manufacturing industries. Nevertheless, the problems with this inspection approach are,
a) AOI can spot the defect but it can not tell you what the defect category is once a defect is spotted and trapped.
b) Human inspection relies heavily on the naked eyes and experience to classify the defect category but this practice is not very reliable (a highly experienced inspector might deliver a higher accuracy rate but he/she could not overcome fatigue), time consuming, labor intensive, and low in overall accuracy rate.
IBM Visual Inspection for Quality (VIQ) is a solution designed to end this low accuracy rate embarrassment in product surface defect inspection and prevent labor waste which it can be applied and used in various manufacturing industries. Basically, VIQ adapts a machine-learning approach to learn the defect pattern with the associated defect type (predefined defect code or category), to trap the defect as it sees any in action, and to classify the spotted defect into a predefined defect category for later analysis in the production line as shown in the following Figure -1 How IBM VIQ Works:

Figure -1 How IBM VIQ works

a) Train
This is the initial learning stage which the purpose of this stage is to construct the knowledge base for VIQ. VIQ must know what the actual defects look like, and what category of errors that the defects should belong to before it can make any judgment on the product in the production line. It is done by learning the defect patterns from actual defect product imagines. In general, more defect imagines VIQ captured and learnt, higher accuracy rate result it can deliver.
b) Deploy
Once goes on production, VIQ uses high-speed cameras to capture product images (or by scanning the images provided by AOI) and then it questions whether the current product under scanned is good or bad. If it is bad, VIQ questions the defect category this bad product should be classified into and labelled it with the accurate defect category.
c) Review, Classify, and Update
If there is any borderline case (the type of defect without a clear presentation of what type of defect it should belong to), or a new type of defect which has never been encountered before, human intervention must kick in. In such a case, the human decision must be made on whether the captured defect should go to any one of the existing defect categories, or a new defect category is needed to accommodate this type of borderline case. It is important that VIQ learns this (either new or existing) type of defect again regardless whether it finally goes to the existing defect category, or classified as a new type of defect.

So, what are the kudos? I can think of the following, at least:
a) Fatigue free – It never gets tired or needs to rest. It works 24×7 without performance degradation, guaranteed.
b) Human error avoidance with high accuracy rate– It has been proven that VIQ delivers around 95% of accuracy rate after a 30,000+-defect-imagine leaning cycle as a result in the display panel manufacturing industry. It is a huge improvement compared to the average 70%+ result in using traditional inspection approach. The high accuracy rate is achieved by using machine-learning approach instead of human or heuristic programming approach for defect detection and classification. In heuristic approach, it is too complex to write proper codes to handle the diversity and fuzzy cases captured, and too difficult to get the judgement and classification correctly to the cases during the image processing steps. On the other hand, AOI+ human approach means more potential human error involvement which is difficult to achieve a high accuracy rate.
c) A better deployment of human resource and cost saving – inspection is one important part of maintaining high product quality but the traditional way of inspection is way too high in cost and low in efficiency due to the massive use of human operators to perform a dull and error-porn job. VIQ can easily replace human workers with a lot higher accuracy rate and speed in inspection result which it implies a huge manufacturing cost saving.
d) Knowledge retention – VIQ captures and maintains the whole product defect knowledge of the manufacturing. It obviously is a physical asset that stays in the FAB forever.
If we take product defect as a crime, the crime must be committed by a chain of events occurred during the process of the manufacturing. Resolving crime of any kind in manufacturing could never be an easy task which it usually requires various tools, approaches, QA measurements, and proper analytic steps. In terms of cracking the crime, VIQ categorizes defects into the evidence and provides the clues in the crime scene in high speed and accuracy that could help finding out who really is behind the smoking gun in your manufacturing. Why don’t you give it a try?

Subject Matter Expert, IBM Global Electronics Center of Competence

More Industry Insights stories

Consortium-supported Blockchain Applications: Anti-Counterfeiting

In an earlier post[1], I described how some excellent use cases for blockchain technology may be held back by either lack of sufficient information on the full cost of process problems outside a company, or by a general preference at companies to not “go it alone” in building a new blockchain solution. The consortium approach […]

Continue reading

Blockchain for the EMS (Electronics Manufacturing Services) Supply Chain

When you’re first exposed to blockchain technology, one question that comes up is where and how to apply it. The answer is often a variation of this phrase: “Blockchain excels at solving issues with ‘high-friction[1],’ multi-party processes where there is a lack of trust.”   The Electronics Manufacturing Services (EMS) industry would appear to check […]

Continue reading

Why use Blockchain instead of another technology?

I seem to get this question every three or four weeks. It’s partly my own fault: when I was first learning about Blockchain, I noticed that many presentations spent 20 or 30 minutes on details of the technology. By the time the topic shifted to how the technology could be applied, most people’s eyes had […]

Continue reading