The history of the manufacturing industry is the history of technological revolutions. New machines, new innovations, and now, ever-increasing computing power continue to change manufacturing. Machine vision, coupled with artificial intelligence and deep learning, is one such innovation.
Industrial technology research firm ABI Research has released a report analysing the uptake of new technology, specifically deep learning, in the manufacturing industries. The report, Machine Vision in Industrial Applications, is part of the company’s artificial intelligence and machine learning research service, which includes research, data, and analyst insights.
Machine vision systems are a staple in production lines for barcode reading, quality control, and inventory management, according to the report. It a mature technology with established incumbents, but advances in chipsets, software, and standards are brining deep learning into the sector. Deep learning is a subset of artificial intelligence where networks and layers of algorithms are created without human intervention. Each layer provides a different interpretation of the data it feeds on.
According to the report, total shipments for machine vision sensors and cameras will reach 16.9m by 2025, creating an installed base of 94m machine-vision systems in industrial manufacturing. Of that, 11 per cent will be deep learning based.
Lian Jye Su, Principal Analyst at ABI Research says these solutions often have long replacement cycles and are less prone to disruption. “Due to the increasing demands for automation, machine vision is finding its way into new applications,” he says.
“Robotics, for example, is a new growth area for machine vision; collaborative robots rely on machine vision for guidance and object classification, while mobile robots rely on machine vision for SLAM [simultaneous localisation and mapping] and safety.”
Deep learning-based machine vision is data-driven and uses a statistical approach in processing the data. This approach allows the machine vision model to improve as more data is gathered for training and testing.
Chipset vendors are launching new chipsets and software stacks to facilitate deep learning-based machine vision. According to the ABI report, Xilinx, a field programmable gated array (FPGA) vendor, partnered closely with camera sensor manufacturer Sony and camera vendors such as Framos and IDS Imaging to incorporate its Versal ACAP System on Chip (SoC).
Intel, on the other hand, offers OpenVINO for developers to deploy pre-trained deep learning-based machine vision models through a common API to deliver inference solutions on various computing architectures. Another FPGA vendor, Lattice Semiconductor, focuses on low-powered artificial intelligence for embedded vision through its senseAI stack, which offers hardware accelerators, software tools, and reference designs. These technology stacks aim to ease development and deployment challenges and create platform stickiness.
On the standards front, vendors are bringing 10GigE (Gigabit Ethernet) and 25GigE cameras into industrial applications. Continual upgrades on video capturing and compression technologies also generate a better image and video quality for deep learning-based machine vision models. This, the ABI report says, ensures the futureproofing of machine vision systems.
“Therefore, when choosing machine vision systems, end implementers need to understand their machine vision requirements, consider integration with their backend system, and identify the right ecosystem partners,” Lian Jye Su says. “Deployment flexibility and future upgradability and scalability will be crucial as machine vision technology continues to evolve and improve.”