ONE WARE redefines traditional industrial automation by introducing quantization-aware AI on the manufacturing line with MAX® 10 devices
The Challenge: Catching Defects Without Draining Power
In today's automated factories, catching production errors immediately is crucial to avoid wasted materials, energy, lost income, and customer dissatisfaction. Artificial Intelligence (AI) offers a powerful solution for automated quality control, improving accuracy and speed. Over time, AI-powered systems develop the ability to identify various types of defects. This self-learning feature makes AI-powered solutions more effective than traditional machine vision systems, particularly in industries where defects are diverse. However, traditional implementations relying on CPUs or GPUs have limitations. Integrating AI on CPU or GPU-based solutions often consumes significant power and is rigidly designed with a single architecture to address multiple use cases. Existing software tools are also limited to universally applicable AIs for training, often overlooking specialized scientific findings tailored to specific applications. This approach can create bottlenecks and increase latency while impeding efficiency and reliability. Additionally, the large space requirement makes it impractical for deployment across numerous inspection points on a factory floor. There's a clear demand for edge AI solutions that are not only fast and accurate but also power- and space-efficient while customizable for specific applications.
ONE WARE is at the forefront of revolutionizing traditional automation through AI powered by FPGAs, with their specialized platform, "ONE AI." Their goal is to save trillions of dollars lost due to quality issues and machine failures. Altera, a longstanding partner, collaborates with ONE WARE as part of the Altera Solution Acceleration Partner (ASAP) Program. ONE WARE has fine-tuned its AI solution to operate on Altera FPGA devices.
The Solution: ONE AI
ONE WARE is tackling the challenge of deploying accurate, flexible, and high-speed AI quality inspection at the edge with its unique ONE AI platform. The ONE AI platform acts as an automated AI developer, simplifying the creation of custom, hardware-tailored AI models using quantization-aware training (QAT). Unlike traditional software tools, ONE AI can use previously overlooked scientific findings related to specific applications to generate Convolutional Neural Network (CNN) based models tackling defect detection, quality control, and root cause analysis.
ONE WARE is the first company worldwide to have developed a system capable of automatically creating fully application-specific neural networks in seconds, tailor-made for each individual task and each target hardware platform. This distinguishes ONE AI from all generic, manually developed architectures and provides significant technological advantages.
This precise specialization means that ONE AI achieves incredible efficiency that is unmatched in power consumption, accuracy, latency, and data requirements. This is especially relevant for edge applications with extremely limited resources.
Since ONE AI adapts individually to each application and hardware, it can be used across all industries, from food production and semiconductors to automotive, medical technology, mechanical engineering, or even aerospace. It can even export these optimized models directly into VHDL using its AI-to-VHDL converter, optimized for Altera devices like MAX 10, Cyclone® 10, Cyclone V, as well as Agilex™ 7, Agilex 5, and Agilex 3 FPGAs and SoCs.
The Result: The Strawberry Test In-Action
ONE WARE demonstrated (watch video) their ONE AI platform by utilizing a VHDPlus Core MAX® 10 FPGA board equipped with a camera. The system inspected strawberries moving via a small conveyor belt. Attendees saw the system instantly identify and flag defective strawberries in real-time, operating at high speed while consuming minimal power. By leveraging the efficiency of MAX® 10 FPGAs, ONE WARE achieved a powerful AI quality inspection solution that consumed ultra-low power, as little as 1 Watt, making it ideal for deploying AI directly on machines. This demo showcased defect detection and other image processing functions running at an impressive 70 frames per second on the MAX® 10 device-based camera, drastically outperforming a standard CPU-based NUC, which managed only ~3 fps on the same task, allowing for real-time inspection on fast-moving lines. The demo contrasted the FPGA-based detection speed with a much slower CPU-based implementation side-by-side, highlighting the performance leap enabled by ONE WARE's optimized AI on the MAX® 10 platform.
"One Ware's ONE AI platform demonstrates how AI inference can be seamlessly integrated into FPGA technology, enabling manufacturers to achieve high accuracy and efficiency without costly hardware upgrades. By leveraging Altera's MAX® 10 FPGAs, ONE WARE accelerates AI adoption in industrial applications, delivering superior performance with 24x higher accuracy and low latency compared to traditional GPU-based systems," said Deepali Trehan, Vice President of Product Management and Marketing at Altera.
ONE AI optimizes models for high accuracy with less data, achieving results like 90% accuracy with surprisingly small training datasets, as few as 60 reference images in the demo, with options for auto-updates to improve further. This combination of optimized AI and efficient FPGAs also leads to reduced TCO and footprint, significantly lowering system complexity and cost. Instead of needing a dedicated PC per camera, fewer central units are needed, saving hardware costs, space, and energy.
In-Depth Whitepaper: Showing Edge implementation of AI-based Defect Detection System using One AI & MAX® 10 compared to GPU solution.
Altera and ONE WARE's joint Whitepaper comparing an FPGA and a GPU-based AI solution for industrial defect detection, reveals that FPGA-based AI solutions for manufacturing factory floor solutions at the edge outperform GPU implementations with 24x higher accuracy, 80x better fps, 20x lower power consumption, and 488x lower latency. Altera’s MAX® 10 FPGAs, known for their low power and real-time processing, are ideal for AI-powered defect detection, offering flexibility and programmability to create custom AI networks. ONE WARE's ONE AI platform enables efficient integration of AI into existing FPGA designs, ensuring high performance, low latency, and significant cost and time-to-market benefits. Read the Whitepaper and see for yourself the impact of ONE AI & Altera FPGAs
Getting Started with ONE WARE and Altera
Altera MAX® 10 and Cyclone® FPGAs provide an excellent, low-power and cost-optimized hardware platform for deploying efficient edge AI applications today. Customers looking to implement high-speed, low-power AI quality inspection, root cause analysis, or predictive maintenance can leverage ONE WARE's ONE AI platform and expertise on Altera devices.
→ Video: See ONE WARE’s live demonstration of ultra-low power AI for quality inspection here
→ Learn more about ONE WARE and the ONE AI platform
→ Stay connected for more partner insights
Subscribe to the ASAP Newsletter for new demos, AI design tips, and FPGA updates. Subscribe here
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.