Image credit: Mariner
Sage Automotive Interiors, based in Greenville, SC, had a challenge. The company supplies fabric and fabric-covered parts to major automakers for use in vehicle production. Sage deployed a machine vision system to ensure it provides quality materials and to identify defects in the textiles it uses.
The machine vision solution could identify inconsistencies in the fabric, but it couldn’t differentiate between lint and stains on the fabric. As a result, the system tagged each possible flaw, and Sage had human inspectors review each one to confirm legitimate defects. In a 1,000-meter roll of fabric, there could be hundreds of potential defects to review. It was costly and inefficient, and Sage wanted to step up its game.
Sage’s existing system required the company to employ a platoon of people tasked with follow up inspections. “It slowed the production line,” says David Dewhirst, vice president of marketing at Mariner. “The problem was impacting their output—cutting their output in half, plus the cost of the human inspectors, all because their [original] machine vision couldn’t do it right.”
Deep Learning Pays Off
Sage turned to Mariner, the developer of Spyglass Visual Inspection (SVI), a deep learning artificial intelligence engine designed for defect detection on the manufacturing floor. Sage decided to deploy SVI as an overlay. Using annotated images and deep learning, Mariner’s data scientists worked with Sage to train the AI models to differentiate between stains and lint.
Sage witnessed a dramatic change after deploying the SVI solution. SVI’s deep learning AI models increased defect detection accuracy by nearly 20 percent to 96 percent. It also reduced false positives from 23 percent to less than 1 percent.
“We built a solution and trained it to detect the actual defects and not to detect the non-defects,” Dewhirst says. “It’s been so effective that those inspectors are doing other things, so Sage was able to shift those costs elsewhere and double their throughput.”
New and Improved Machine Vision
Manufacturers can spend 40 percent of their annual revenue-producing defective products. Older machine learning systems tend to be more programmable, assessing the shape of a defect rather than its specific features, such as a fuzzy edge.
“We’re tackling the most difficult defect detection problems,” says Phil Morris, co-founder and CEO of Mariner.
This image from Mariner is an example of defective and non-defective product pulled from Sage Automotive Interiors. Images #2 and #4 show defects.
SVI can operate over an existing machine vision solution. It sends data to an edge server for on-premises inferencing and AI-driven decisioning. The edge device forwards the relevant data to the cloud for additional processing and retraining of the AI model.
The Sage installation proved that a next-gen machine vision solution could drastically impact a company’s bottom line. When SVI identifies the potential source of a problem, the system sends a notification to employees. It can correlate data from other sources in the factory, enabling it to perform root cause analysis.
“If we solved this problem for one provider, we can find every customer using those older systems and offer the same solution,” says Morris. In fact, the Sage deployment was so successful that Milliken, its former parent company, decided to install SVI across multiple locations.
Optimal Performance Through Partnerships
The SVI solution works with Intel® Xeon®-powered Dell edge servers provided by Arrow Electronics and Microsoft Azure. Mariner partners with North Coast Technical Sales, a system integrator based near Cleveland, Ohio, if a customer needs cameras and lighting.
“The global presence of Arrow is huge. Our objective is to grow our global network of resellers,” Morris says.
- Learn more about Mariner.
- Find out more about Arrow Electronics.
- Learn more about North Coast Technical Sales.
- Discover more about the Intel® Xeon® processors.
- See the Mariner Spyglass in the Intel® IoT Solutions Marketplace.