IoT video analytics & quality control at the edge

[IoTExperts]

Steve Fearn,

Chief Technologist, HPE Pointnext

Automated assembly lines have been a part of modern factories for decades, with machines handling practically every stage of the manufacturing process. However, one step that has remained within the domain of human overseers is quality control. It’s a critical task, but it is also becoming more difficult. As product refresh cycles shrink, components reduce in size, designs are updated more frequently, batch sizes become smaller and customers demand more customizations at the point of manufacture, effective quality control is becoming a huge challenge for progressive manufacturers. 

That is starting to change, though, thanks to the IoT technology revolution sweeping across industry. With new video analytics systems that combine video sensors, machine learning, and high-powered edge processing, manufacturers can speed production, improve quality control, and have human workers concentrate on higher-level tasks. 

Video sensors bring precision and speed 

The concept behind video analytics for quality control is simple. Production works basically as before, using the same legacy equipment, but with video sensors placed at inspection points along the line. The video feeds are sent to nearby edge hardware to be processed. There, the video analytics application compares the video feed with details from the plant’s MES or a bill of materials, and ensures that the finished products match the build list. 

The video analytics system can check for various flaws, ranging from scratches to misaligned parts. It can also spot problems that are particularly difficult for human inspectors to gauge. For instance, at a factory that makes computer servers, the sensors can check the serial numbers of memory DIMMs and dozens of other components that are installed onto a motherboard chassis. Not only can the system verify that the tiny pieces were mounted properly, it can also make sure that the correct components were used. This degree of precision is impractical for high-volume manufacturing using manual QC inspections. 

Even better, the system’s machine learning capabilities mean that accuracy will improve over time. During the initial training phase, ML algorithms may require many thousands of images to determine what can pass the QC protocol. Afterwards, the system can use new data to update its models, leading to lower error rates. 

The edge factor 

Having processing take place right on the factory floor is crucial. If the system has scores of cameras generating thousands of images every hour, uploading everything to the cloud would entail massive bandwidth costs, not to mention latency and other inefficiencies associated with sending data up to the cloud and back down to the plant. Further, no one wants to deal with the risk of disruptions between the plant infrastructure and the cloud -- the whole line would just stop. 

Fortunately, HPE Edgeline system is designed for this type of high-performance computing scenario. The ruggedized devices can host the video analytics application, and process the incoming feeds to compare the images from the production line with reference photos and other data. When you’re looking at multiple production lines with hundreds of cameras as well as other types of industrial IoT sensors, a device like the Edgeline may be the only realistic option to run powerful analytics applications at scale.