Edge AI drives smart manufacturing

By Gina Roos, editor-in-chief

Smart manufacturing uses a number of technologies, including digitalization, internet of things (IoT), big data, and robotics. To increase information visibility and system control in manufacturing systems, manufacturers are deploying advanced sensors and control systems with artificial intelligence (AI) algorithms.

The global market for smart manufacturing is forecast to reach $370 billion in 2022, growing at a compound annual growth rate of 10.7%, according to TrendForce . What makes this market so large are the use cases. Smart manufacturing encompasses a variety of applications that include smart factories, smart supply chains, automated delivery vehicles, and robotic arms.

The global number of AI edge devices is forecast to jump from 161.4 million units in 2018 to 2.6 billion units by 2025, according to Tractica LLC . The top AI-enabled edge devices include mobile phones, smart speakers, PCs/tablets, head-mounted displays, automotive, drones, robots, and security cameras.

Over the past few years, the Hannover Fair  has been a showcase of what’s to come in Industry 4.0 — collaborative robots  (cobots), digital twins , mixed reality , predictive maintenance, drones, and AI applications. Some of the biggest players, according to TrendForce, are Universal Robots, Siemens, STMicroelectronics, Xilinx, and GE.

As smart manufacturing opens the floodgates to solutions for handling huge amounts of data, latency, and bandwidth costs, edge computing is predicted to become the cornerstone of predictive maintenance, according to P.K. Tseng, analyst, TrendForce. “Big data, precision analytics, and higher-performance hardware have been the three big driving forces pushing AI from the cloud down to end devices and encouraging the combination of edge computing and AI.”

As manufacturing operations continue to add AI, they are getting much more granular with the data they are collecting, the processes they are monitoring, and the identification of patterns to help them see if there are anomalies, defects, or any sort of issues, said Keith Kirkpatrick, principal analyst, Tractica. “The big problem with the way an AI solution is deployed over a centralized platform is the latency issue between the particular machine and the system. So, the idea of incorporating AI at the edge is to eliminate that latency. So if some sort of anomaly is detected, it can be corrected almost immediately, and that is particularly important as we get into manufacturing with very precise tolerances.” 

In addition to latency issues, network traffic is a big issue in manufacturing or production systems. “It’s not that they just have a couple sensors and couple of edge devices,” said Kirkpatrick. “They are deploying this technology across their entire production networks, and with that you’re seeing a lot of traffic going across that network, and it’s just not data; there are also video streams, particularly if they are using it to conduct final inspections or making sure the final products meet certain quality standards. Off-loading some of that data off of the network and processing it at the edge is a real benefit.”

“The added value from AI will boost that of edge computing in Industry 4.0 in the four major areas of real-time decision making, cost reduction, operation reliability, and security enhancement, and will receive even wider adoption,” Tseng said. “The main challenge lies in whether AI algorithms will continue to advance and increase in precision, as well as whether companies may be able to bear the costs from software and hardware upgrades.”

Blending smart manufacturing with edge AI delivers several benefits, including real-time decision-making, lower costs, operational reliability, and enhanced safety, said Tseng.

“Edge AI allows end equipment to retain some decision-making ability and gives immediate responses without constant connection, circumventing the need to transfer everything to the cloud,” said Tseng. “Thus, it does save bandwidth cost and power consumption to a certain degree.”

However, Tseng said, the biggest strength of smart manufacturing with edge AI is increased operational reliability.

AI providers

Chipmakers and cloud providers are both playing a big role in edge computing. Key players include chip giants such as Nvidia, Intel, Qualcomm, and NXP, and leading cloud providers such as AWS, Google, and Microsoft. But TrendForce believes Taiwan-based chip suppliers have a good shot at a slice of the edge AI market thanks to government-granted resources they can access.

“AI chips may be roughly categorized into various different structures, such as CPUs, GPUs, FPGAs, ASICs. Compared to CPUs and GPUs, over which giants have a monopoly, logic chips used in AI edge computing as well as FPGAs and ASICs, which are mainly used for imaging, visualization, or customization, may afford a better chance of development for Taiwan,” said Tseng.

Tractica predicts that the industry will start to incorporate more ASICs, which are self-contained chips, that will be programmed to handle very specific tasks, said Kirkpatrick. “The idea is that it will be relatively lower power; it will have the processing power to handle that particular task, but it’s not going to be designed to handle everything, such as a more centralized processor like a GPU because some of the GPUs are extremely energy-intensive.

“The challenge on the edge is you need to have a processor that is powerful but also one that won’t generate too much heat or drain too much power because that really impacts the overall temperature of the system,” explained Kirkpatrick. “If it gets too high, it could throw off the production processes, and when talking about power, there is a cost attached to that as well. It may not be so much on the individual machine or sensor, but if you multiply that by hundreds or thousands of sensors, it’s a cost that manufacturers need to take a look at.”

Industry 4.0 has always been pushing businesses to embrace digitalization, with other technologies including IoT, big data, and robotics, on the road to smart manufacturing, but it’s a time-consuming and costly transition for companies, said Tseng, whether they are deploying industrial IoT, introducing smart manufacturing, or constructing smart factories.

Tseng believes that many traditional manufacturing industries that lack digital roots can move to smart manufacturing by introducing digital tools and integrating different fields of industry, enabling them to enter the supply chain of industry giants or collaborate with them.

“Companies are in the process of digital transformation, requiring the introduction of cloud or IIoT [industrial IoT] platforms, data analysis and management, hardware/software integration, and other technologies, which will allow them to work with the cloud industry, system integrators, telecommunications, etc.,” said Tseng.

“Furthermore, suppliers, whether at the upper or lower end of the supply chain, have to undergo horizontal integration,” he added. “Machinery manufacturers may, for example, release knowledge on processes to the public. They may then provide designs for robotic arms and prototypes, and suppliers of industrial computers may come up with an integrated solution to link everything together. This cooperation among heterogeneous industries thus constitutes a complete smart manufacturing process.”