What Manufacturers Should Know When Implementing Edge AI

Vice President Network and Edge Group, General Manager Federal & Industrial Solutions at Intel Corporation.

Artificial intelligence (AI) continues to be a topic in both the public and private sectors. While it feels like a “new” technology, the manufacturing sector has leveraged AI in multiple ways for some time. Tasks like predictive maintenance, analytics, quality inspections and demand forecasting systems all use AI.

Until now, these AI applications have been primarily handled in the cloud. But with the continual evolution of edge computing, there is a push to run inference directly on data closer to where it’s generated—on the factory floor.

Made possible using common off-the-shelf technology and rapidly evolving software, edge AI can help enable manufacturers to make near-real-time, informed decisions that allow operations to be more efficient at a lower cost and eventually “close the loop” to run autonomously. Let’s take a look at where edge AI is today, obstacles to its full deployment and approaches that manufacturers can take to address implementation challenges.

Staying relevant in our current economy requires agile business strategies that drive innovation while keeping costs under control. Automation running on edge devices is part of the key to meeting these requirements:

Manufacturing often relies on latency-critical applications, some with sub-10 millisecond response times. AI operating at the edge has the responsiveness needed to help ensure effective collaboration between machines and humans. This near-real-time processing enables the ability to make fast, data-driven decisions that directly impact production as it is happening.

Sensors, video cameras, IoT devices and machines are creating more data than there is bandwidth to transmit. Edge devices bypass the limitations of bandwidth by having the necessary AI needed to inform their autonomous operations. By adopting edge computing, manufacturers can cut the financial cost and time spent on addressing bandwidth concerns.

Traditionally, devices communicating to the cloud present potential security risks. Since edge devices have advanced AI capabilities built on top of robust security protocols, they can address and protect sensitive data and meet global data sovereignty, compliance and privacy regulations.

The imperative for greater agility and control has accelerated edge AI application development. I’ve seen firsthand future-focused manufacturers continue to demonstrate the advantages of creating seamless interactions between humans and machines with:

• Optimized factory floor layouts that use AI at the edge to create flexible processes that adapt to changing market conditions within days rather than months.

• Autonomous moving robots that use edge computing to navigate warehouses in real time. By doing the heavy lifting and transportation of goods, they free up workers for more high-value tasks.

• Real-time augmented reality is made possible by running AI at the edge, enabling it to guide workers and improve safety, efficiency and precision.

• “Factory-in-a-box” systems that integrate software, machinery, digital twins, ordering systems and analytics into an ecosystem on the edge so they can support real-time insights and decision-making.

Though estimates vary, over half of all AI projects fail. Edge AI has specific hurdles:

Making accurate decisions in AI initiatives is crucial, so accessing high-quality data is critical for successful AI projects. In manufacturing, data is diverse and complex. If it can be accessed, it is often outdated, error-riddled or siloed. This places a greater reliance on human knowledge to build AI models. However, these experts typically lack data science skill sets. The scarcity of experienced data scientists and AI professionals further complicates the task.

Edge AI projects often occur in brownfield environments with proprietary equipment lacking connectivity and intelligence. Limited space, power, connectivity and environmental obstacles make implementation difficult. Diverse machines and production systems use various technologies, while outdated software compatibility issues hinder scalability and accurate model deployment in production settings. Many AI models struggle to compress large AI models onto small devices in extreme environments.

Security is a primary concern when dealing with distributed edge infrastructure and devices outside centralized data centers. The complexity of securing digital and physical environments increases with edge deployments. Edge management is complex, with heterogeneous nodes, operating systems, applications, compliance restrictions and computing architectures. Traditional cloud technology does not seamlessly extend to the edge, necessitating new tooling and platforms.

Overcoming these hurdles to drive successful business outcomes requires new approaches to handling data, environments, and users that successfully blend OT and IT systems.

A software-defined approach addresses the challenges of implementing edge AI in factories. Transitioning from fixed-function hardware to software-defined systems enables manufacturers to work in time-critical environments. This allows them to address issues caused by diverse data types, power, performance, latency and cost, even in extreme conditions.

Open standards and platforms are the foundation for merging OT data from legacy devices into AI-based IT. Beyond the immediate benefits of better data access and analytics, companies position themselves to incorporate new technologies quickly and efficiently when they embrace a software-defined methodology.

Edge AI performs real-time processing and analytics on the shop floor, while cloud AI can perform deeper analysis and learning on aggregated data. Organizations can tap into the best of both worlds when their ecosystems encompass cloud, PC or edge environments.

Nontechnical domain experts possess critical insight into industry-specific models. IT departments can choose software applications that are much more accessible to these experts.

Data from edge devices needs to be secure at rest, in transit and in use. Software and high-performance silicon found in newer IoT devices provide deep and extensive defenses for protecting data in zero-trust networks. This type of security is not inherent in fixed-function hardware.

The challenges manufacturers face today can be addressed by adopting a software-defined approach that allows for adaptive, connected, industrial systems capable of learning while supporting flexible workloads. Organizations willing to embrace AI at the edge better position themselves to reap the business advantages of intelligent industrial systems.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?

Alibaba Puts Heat On AWS, Huawei With New Chips, Servers, Data Centers

The cloud computing giant is stepping up its rivalry against cloud competitors with the launch of new in-house processors, servers and new data centers outside China.

Alibaba is stepping up its game against cloud competitors including Amazon Web Services and Huawei by developing new in-house processors and a line of servers to boost its cloud computing data centers, while also unveiling plans to build new data centers outside China. The new server and semiconductor innovation is aimed at boosting the China-based cloud giant’s computing capabilities and global market reach against AWS and Huawei, which both have also released server chips of their own.

Alibaba’s new chip, dubbed the Yitian 710, is built on 5nm process technology and is powered by 128 Arm cores with a 3.2GHz top clock speed. Alibaba said the Yitian 710 is the first server processor that is compatible with the latest Armv9 architecture.

[Related: Gartner’s Top 5 Biggest IT Sales Growth Markets For 2022]

“Customizing our own server chips is consistent with our ongoing efforts toward boosting our computing capabilities with better performance and improved energy efficiency,” said Jeff Zhang, president of Alibaba Cloud Intelligence, during Alibaba’s annual Apsara Conference this week. “We plan to use the chips to support current and future businesses across the Alibaba Group ecosystem. We will also offer our clients next-generation computing services powered by the new chip-powered servers in the near future.”

The Yitian 710 server chips aren’t commercially available and will run in proprietary, cloud-native servers. Alibaba is developing a new line of proprietary servers, dubbed Panjiu, which will be powered by its new Yitian 710 processors. The combination of new chips and servers is a move to boost Alibaba’s overall cloud services by optimizing energy efficiency and computing performance.

By separating computing from storage, Alibaba said its Panjiu servers are optimized for general-purpose and specialized artificial intelligence computing on Alibaba’s cloud, as well as high-performance storage.

The Panjiu servers have a modular design approach for large-scale data center deployment as well as for a wide variety of cloud-native workloads, such as containerized applications and compute-optimized workloads. The company did not provide a time frame as to when the new proprietary servers will be in production inside Alibaba’s data centers.

Alibaba is the leading cloud computing company in China but is now focusing on growing its footprint in other Asia geographies.

This week, the company unveiled its plan to open its first data centers in South Korea and Thailand. The new data centers will be opened sometime in 2022, according to Alibaba.

“We are committed to bringing our world-class, hyper-scale cloud infrastructure to Korea and Thailand, so that we can support local businesses to be at the forefront of digital transformation,” said Selina Yuan, general manager of international business, Alibaba Cloud Intelligence, in a statement.

Earlier this year, Alibaba opened or unveiled plans to open several new data centers in the Philippines and Indonesia.

CBSE Board Exam Date Sheet 2024: Timetable for Classes 10th and 12th released; Download here

CBSE Board Exam 2024: CBSE Class 10 Date Sheet for Important Subjects

CBSE Class 12th Date Sheet 2024: Science Stream (Main Subjects)

CBSE 12th Date Sheet 2024: Commerce Stream (Main Subjects)

CBSE Class 12th Date Sheet 2024: Humanities (Arts) Stream (Main Subjects)