About Elementary

• US-based organization, creating AI-powered machine vision solutions for manufacturing.
• Focused on improving efficiency and reducing waste in manufacturing.
• Pioneers of VisionStream, an AI model that catches up to 99.99% of defects.

Overview

Manufacturers need fast, securely designed, and reliable machine vision to keep production lines running. Elementary is a pioneer in AI-powered visual inspection, differentiated by their cloud-compatible system. When it began scaling its edge devices, traditional server-based architectures no longer worked: they carried undesirable security risks, and were time- and effort-intensive. By migrating to Canonical’s Ubuntu Core, supported with NVIDIA and Intel® GPUs, Elementary streamlined provisioning from a full day to just 20 minutes. They gained robust security features like full-disk encryption and secure boot, and enabled customers and partners to self-manage their devices at scale. Today, Elementary solutions deliver over 1 billion inspections per year.

Challenge

In 2017, Elementary was founded to transform manufacturing quality control with AI-powered computer vision. Before AI-powered systems were a reality, traditional machine vision systems were the go-to technique for quality control. While these machine vision systems were useful, they were also brittle. They relied on rigid, rules-based setups and often failed to function in real-world conditions – for example, when there were changes in light levels or variations in the products being inspected. These minor variations in conditions often resulted in the system becoming misconfigured, requiring a specialist in machine vision to reconfigure the system manually. As Peter Lipay, Vice President of Engineering at Elementary, explained: “We've often found we’re not the first machine vision system that our customers are deploying. They have one sitting there, unplugged, because it stopped working at some point. They just stopped using it, because it was too much of a pain to keep it in sync with what they were doing.”

Elementary’s approach has significantly improved machine vision systems. Its AI models can be trained on real-world data, and are flexible enough to adapt to changes in the operating environment or working conditions. For one of Elementary's Fortune 500 customers, this reduced defect rates from 24% to 1%. Additionally, IoT connectivity enables cloud analytics that provide greater oversight and control for manufacturers. However, as Elementary scaled, it outgrew its infrastructure. Running traditional server-based architecture on edge devices meant Elementary carried the heavy burden of OS updates, provisioning, and field support. Updates, configuration, and provisioning all required physical access to devices, increasing operational expenditure and creating data security concerns.

Elementary’s machine vision inspection station on a factory floor

Solution

Elementary evaluated a wide range of alternatives, from engineering the solution themselves to third-party solutions like the Yocto Project and others. The solution had to be compatible with Ubuntu Server, or be straightforward enough to make migration of current workloads feasible. Elementary also took into account compatibility with their pre-existing hardware architecture. Elementary uses NVIDIA and Intel® GPUs – their large-form factor devices use discrete NVIDIA GPUs, whilst their small-form factor devices use integrated Intel® GPUs. Both their large-form factor and small-form factor devices use Intel® x86 CPUs.  Elementary’s architecture also includes the ASRock Industrial iEP-7020E industrial Internet of Things (IoT) controller, which features 13th Gen Intel® Core™ i7-1365URE processor, and the Dell Precision 3680 tower workstation, which features the 14th Gen Intel® Core™ i7-14700 processor.  Dell’s commitment to producing high-quality hardware, and Intel’s compatibility with Open Neural Network Exchange (ONNX) format machine-learning models, meant that Elementary wanted to find a solution that would work with this hardware.

Most solutions Elementary considered either lacked critical features such as atomic OTA updates, full-disk encryption, and secure boot, or carried prohibitive engineering and cost burdens. Following testing several solutions with a proof of concept, Ubuntu Core emerged as the only option that combined security, low-touch management, long-term support, and cost efficiency in one platform.

Results

Migrating to Ubuntu Core has unlocked secure, scalable global deployments for Elementary, and freed them from the limitations of server-based architectures. 

Provisioning – once a full-day manual process – now takes 20 minutes. Devices can be securely factory reset and activated directly by customers or partners, eliminating the bottleneck of relying on Elementary for setup.

Elementary’s machine vision inspection system. From left to right: the Shroud front view (with integrated camera, lens, light, trigger, and enclosure), the Shroud side view, the Vertex AI controller, and the Radian camera, with Node attached.