Smart Manufacturing: What’s Needed For The Industrial Intelligence Revolution?

Smart manufacturing – the use of nascent technology within the industrial Internet of things (IIoT) to address traditional manufacturing challenges – is leading a supply chain revolution, resulting in smart, connected, and intelligent environments, capable of self-operation and self-healing.

While factory automation has been around for decades, smart manufacturing goes a great deal further. It uses a combination of artificial intelligence (AI), robotics, digital twins, additive manufacturing, and powerful cloud-based compute to deliver new levels of flexibility and intelligence.

Smart manufacturing is a key element of ‘Industry 4.0’, or the fourth industrial revolution. It enables companies in all corners of the manufacturing industry to move beyond traditional benefits – uptime and strength – to focus on quality, human productivity, and overall factory efficiency. The result is increased profit due to improved throughput, higher yield, and reduced waste.

These technologies include the sensors that generate huge amounts of data, the data centers that analyze that data, and the control systems for the sophisticated machines that handle, bend, weld, solder, and print the products of tomorrow.

And rather than replacing human workforces, smart manufacturing enables far greater autonomy between humans and these machines. ‘Cobots’ now work in close proximity to humans, bounded by comprehensive functional safety protocols that ensure they keep their softer counterparts safe and enable them to focus on quality, productivity, and higher-level tasks.

Smart manufacturing in silicon

While it might be easier to perceive the benefits of smart manufacturing within the operations of a car manufacturer or grocery store, smart manufacturing also provides a wealth of benefits to high-tech industries.

Semiconductor devices, made from billions of transistors, continue to grow in complexity. In turn, the manufacturing process is becoming ever more complex, and the risk of failure is remarkably high. Semiconductor manufacturers rely on smart manufacturing processes to produce higher yields and achieve higher margins. Data analytics using AI can result in faster failure analysis and other production efficiencies.

Semiconductor fabrication plants, or fabs, cost billions of dollars to build and maintain – which is why there are relatively few of them in the world. Much of that cost goes on equipment, the maintenance of which is vital to ongoing operation. By using smart manufacturing technologies to monitor equipment health and perform predictive maintenance, fabs can reduce unplanned maintenance time significantly.

The global supply chain issues of recent years have also required new approaches to unplanned downtime – and smart manufacturing extends to predictive supply chain management to look ahead and identify coming issues and ways to mitigate them. Many of these technologies played a key role in keeping fabs operational during the 2020-2021 global chip shortage.

Smart manufacturing requires local data processing

Central to all of this is data – and lots of it. A smart factory can generate upwards of 5 petabytes (PB) per week and all of this data needs to be transmitted, stored, and analyzed.

Analytics has of course long been used to optimize the performance of systems. But the original realm of analytics software was procedural and algorithmic, following strategies conceived by MBA graduates and software engineers.

With smart manufacturing, the number of data points expands exponentially to the point at which the procedural approach breaks down. The advent of machine learning (ML) now allows factories to analyze patterns in very large datasets, which are well suited to the kind of massive data analytics common in Industry 4.0.

But this data becomes problematic in itself. How can we handle the transmission and analysis of this incredible amount of information? For example, computer vision is thought to be essential to smart manufacturing in observing the many details of operations within a factory. However, each smart camera might generate thousands of gigabytes of high-resolution video data per day by itself.

Edge data centers rise to the challenge

Sending this amount of data upstream to the cloud is impractical as it overwhelms data networks and creates bottlenecks – especially in real-time computing where the value of data can be measured in milliseconds and any delay or latency in extracting insight reduces the value of that insight to zero.

As a result, dedicated compute facilities, in the form of industrial edge data centers, are becoming commonplace in smart manufacturing. Residing within relatively close physical proximity to where the data is generated, edge data centers reduce latency to near-zero and retain that critical time-to-value while maximizing data privacy and reducing energy costs.

Adding intelligence to sensors

The sensors within factories are becoming more intelligent, too. Low-power machine learning (ML) within each endpoint device is capable of analyzing the data they are collecting and reducing the amount of data transmitted by only sending back the inference.

To use the computer vision example, imagine a smart camera trained to detect physical faults in parts rolling off a production line. Rather than sending every bit of video data upstream to a cloud or edge data center, the device could analyze its own data and only send sections of video that appear to reveal faulty components to the cloud.

And by combining that data with other sensor data through sensor fusion, we can gain deeper insight or achieve new levels of autonomy. For example, adding RFID to each of the items as they are produced might enable individual parts to be flagged for human examination.

With every individual sensor or system aligned via protocols and working together, a factory powered by smart manufacturing becomes far more than the sum of its parts – it is one intelligent entity, optimized together for performance, productivity, and community and with rapidly-interpreted data as its lifeblood.