INDUSTRIAL

For too long, manufacturers have been throwing good money after bad. Having a defective product roll down the production line and pass routine visual inspection only to have it be returned later for rework or refund is expensive and wasteful. And destructive testing of one component out of 100 hardly inspires trust that the entire lot is defect-free.

Instead of a patchwork postmortem inspection process, manufacturing is turning to computer vision solutions that can catch problems the human eye might miss. Despite the increasing popularity of these methods, they are not without their own challenges, says Jonathan Weiss, Chief Revenue Officer at Eigen Innovations Inc., developer of industrial machine vision solutions.

For one thing, many machine vision solutions don’t play well with in-house software, so they are siloed. And often the algorithms address only one specific use case in a closed system, which limits their implementation for other applications.

To address the problems associated with inspection, manufacturers need adaptable, vendor-agnostic computer vision solutions. Deployed inline, they can detect defects right at production time before these problems snowball into larger headaches. The Eigen centrally managed OneView Machine Vision Software makes it possible.

Such inline inspection forms the backbone of the Eigen OneView Quality Inspection for Metals solution, managed by OneView Software, and primarily leans on thermal cameras and uses machine learning models to understand what heat signatures look like for correctly executed industry processes. Working with such a knowledge base helps OneView detect problems happening in real time, when metals are welded, plastics extruded, or when materials go through a range of manufacturing steps.

Detect problems happening in real time, when metals are welded, plastics extruded, or when materials go through a range of #manufacturing steps. @EigenInnovation via @insightdottech

Industrial Applications for AI-based Inline Inspections

Eigen customers can adapt the solution to a number of related applications like injection molding, welding, or adhesive manufacturing processes. In each case, factory teams use OneView to build AI and ML models that learn different types of inspection paradigms.

Case in point: Henderson Stamping, a Tennessee-based manufacturer, struggled with accurate defect detection in components it produced for Whirlpool. The very thin shiny surface film that helped safeguard parts from scratches and dents also prevented thorough manual inspection. As a result, a small but significant percentage of components that were being shipped turned out to be defective. “This can become very problematic for manufacturers who have agreements with their customers, where fines can be imposed for shipping defective goods,” Weiss says.

Eigen helped the company develop a custom inspection solution that leverages the principle of deflectometry. The procedure involves shining light on the surface of the metal and looking for surface defects by evaluating the resulting light patterns. Henderson now inspects all of its components using the OneView managed solution and has significantly reduced OEM recall rates.

Similarly, a manufacturer of large metal grates wanted to ensure its welds were strong enough. Post-production testing involved putting the grates through a torque machine that applied pressure to find weak points. Using OneView software and multiple thermal cameras, the manufacturer can conduct inline testing of all weld points at every single cross section. The software stitches multiple camera images together to create one composite image and pinpoint problems.

The size of the defect that can be detected depends on the sensitivity of the cameras used, but in most cases, those that are a millimeter or larger are a slam dunk, Weiss says.

Computer Vision Leads to Operational Efficiencies

OneView is about more than just detecting defects. “We take it a step further and also show process data. So we help manufacturers not just see that there’s a defect visually, but we also ultimately help them understand the root cause. Not only are we telling them they have a bad product, but we’re also showing them exactly what shifted or drifted in the process that an engineer now needs to fine-tune,” Weiss says.

OneView provides complete traceability so manufacturers can mitigate warranty claims and find a variety of applications for both cost savings and improved efficiency and customer satisfaction.

There’s also a sustainability advantage in detecting defects inline. Shipping faulty products only to have customers return them increases the associated carbon footprint. Catching problems early on in the manufacturing cycle leads to less carbon waste, too. “We’ve developed complete case studies just on the CO₂ footprint reduction that we’ve helped companies with, and it actually extends well beyond the footprint of the factory,” says Weiss. “You’re talking about hundreds of thousands of tons of CO₂ essentially that can be saved depending on the production footprint.”

Open Technology and Tools Enable Flexible Deployments

Inline defect detection comes down to work that must be done in a matter of seconds, which is why Intel technology is especially important to vision solutions designed and managed with OneView. Those time constraints can be a significant challenge, and the Eigen team found that the Intel^® OpenVINO^™ toolkit helped it achieve the speed needed to operate. The performance that OpenVINO unlocks and the speed at which it can inference images is one reason why Eigen includes Intel hardware and software as a “core part” of its technology.

In addition, Intel helps Eigen hit its differentiation metric in being able to deliver flexible deployment options. “We want to be as hardware-agnostic as possible when we provide our solutions, and so OpenVINO became a key part of our architecture because it allows us to support a very wide range of hardware options,” Weiss says.

Eigen has an internal engineering services group that sometimes functions as a systems integrator, although it also works with a network of preferred systems integrators. Eigen works with SIs who implement the solution blueprint that the company draws for clients. Collaborating with SIs is a key part of the company’s strategy as it helps unlock deployment scale—especially for larger customers.

A Must-Have for the Future of Industrial Automation

In the future, expect these machine learning models to get more accurate and deliver better results with fewer training images.

“Our sweet spot is helping folks use thermal applications to see what they otherwise can’t see,” Weiss says. A whole range of processes in a whole range of specialty industries qualify.

The future, Weiss forecasts, will move AI and computer vision-powered inline inspection to a must-have instead of a nice-to-have. Using such inspection tools also helps manufacturers decrease workforce turnover rates as employees now need to understand machine readings instead of visually inspecting products.

The decrease in waste and the cost savings delivered make such solutions a no-brainer, and no longer will manufacturers throw good money after bad.
 

This article was edited by Georganne Benesch, Editorial Director for insight.tech.