October 13, 2021
AI and CV Transform Quality Automation on the Factory Floor
'This market study was conducted together with ABI Research''
This article first appeared on insight.tech on May 27th, 2021
If you’ve ever purchased a defective product, you know how frustrating it can be. You probably wondered, “How can this happen?” For decades, quality control has been an unsophisticated process, typically involving QC specialists performing spot checks of products on a factory line or warehouse floor. Due to human error, faulty units make it out the door and into the hands of customers. The result? Costly waste, warranty claims, and dings on the company’s reputation.
“Checking components is not an easy job,” explains Serhan Can, Director of AI at Relimetrics, a computer vision (CV) and machine learning (ML) software provider. “Human operators get tired and start missing defects. Plus, there’s a huge time pressure on manufacturers and logistics companies that need to make and ship goods. The quality control process needs to take as little time as possible.”
While machines can do repetitive work like inspecting parts, their capabilities haven’t yet been used to the fullest due to complexity and expense. But today, new tools can simplify the process.
Evolution from QC to QA
Moving from quality control (QC) to quality automation (QA) can help companies get to zero defects, but the transition hasn’t happened overnight. Over time the technology has matured and the latest iterations leverage AI to detect anomalies and adapt to production variabilities in real-time.
To understand the impact these new quality control systems can have and where your company falls on the automation progression, consider this QA maturity model with five distinct levels, defined by Relimetrics and ABI Research:
- Level 1: Humans collect and assess data.
- Level 2: Cameras collect data and humans assess it.
- Level 3: Cameras collect data, traditional CV software identifies issues, and humans resolve problems.
- Level 4: Cameras integrated with AI-based machine vision software collect data and identify issues, and humans resolve problems including false detections by the AI.
- Level 5: QA is fully digitized and automated, with human involvement only in marginal cases. Cameras integrated with AI-based machine vision software identify problems and instruct the Programmable Logic Controller (PLC) to scrap the item or send it to a rework station.
Each step lessens reliance on humans. “AI automation offers a greater degree of accuracy than people performing the tedious work,” says Dr. Kemal Levi, Founder, and CEO of Relimetrics. “Most manufacturers are in the beginning of their journey. The majority are operating at Level 3, and very few are operating at Level 5.”
Over the next five years, the rate of QA automation is expected to rapidly accelerate with CV playing a central role. Part of a fast adoption is the availability of new tools. In the past, sophisticated AI technology required high-level coding skills, but today’s systems, such as Relimetrics AI-based QA Automation Solution for Electronics Assembly (RELI-QA), can be self-deployed without any programming or deep learning expertise required.
“Rather than humans writing the software, our software writes itself,” says Can. “The product enables people who have no experience in deep learning to be able to train deep-learning models, automating the whole process.”
Quality Automation in Action
Relimetrics recently helped HPE hardware manufacturer Foxconn get to Level 5 on the maturity model. Using RELI-QA, it automated the QC process for Foxconn’s production of complex HPE servers, which can come with up to 16 memory models, each having 16, 32, 64, or 128 gigabytes. The memory configuration is one of 20 different variables. In addition, the pace of production is high.
“The challenge was that every server is manufactured according to a specific end user’s need, and each server coming down the line is different,” says Can. “It’s a very complex case for a human operator. Checking for defects can take a person up to five minutes per server.”
With RELI-QA, auditing time is reduced to about 30 seconds. In addition to time savings, the QA process reduced the number of defective HPE servers that reached customers by 25%. And the overall production performance went from sigma 2.1 to sigma 4.2.
DL and CV: The Path to Level 5
Powered by Intel® processor-based architecture, RELI-QA uses high-definition cameras at the edge. The solution analyzes and inspects products as they travel through a production line or warehouse. The video stream is transferred to an embedded or attached IT system, where data is compared to the Manufacturing Execution System (MES) defined during the build process. The Intel® Distribution of OpenVINO™ Toolkit optimizes the inference time of the models. If a defect or imperfection is detected, an alert is sent in real-time, fully digitizing the QA inspection process.
“The beauty of this technique is that it learns from images,” says George Sakr, Ph.D., Relimetrics’ deep-learning expert. “It looks at images, extracts the important features, recognizes what differentiates the defected image from a non-defective image, and learns by example. This capability is why AI is leading the transition.”
AI-driven QA creates an ecosystem with feedback loops that provide insights that manufacturers and logistics providers can use to improve efficiency and operations—helping them ship products to customers defect-free.
And in the supply chain, QA automation assures traceability. In the case of a product recall, for example, a company can take quick action and identify affected items as opposed to disposing of an entire batch, generating significant savings.
Closing the loop in production is critical for Industry 4.0, and real-time feedback is how it can happen.
“Instead of waiting until the end of the production to assess whether products have been properly manufactured, any issues can be identified and corrected at the time of manufacturing before they end up in the customers’ hands,” says Levi. “Continuous feedback gives us hope for more efficient processes and higher profitability for the future.”