Artificial intelligence has a big role to play in driving the digital transformation of industrial manufacturing and delivering the benefits of Industry 4.0. Two key applications for AI in manufacturing include predictive maintenance and machine vision for defect inspection and product quality. AI also offers the potential to deliver benefits throughout the product life cycle, ranging from generative design for product development to after-sales and warranty support.
However, there are a host of challenges preventing AI from making inroads in the industrial sector. Some of these roadblocks are simply due to the makeup of the industrial sector with a variety of use cases, company sizes, and lack of expertise, which adds to the complexity of implementation.
There is great interest in AI and machine learning (ML), but there is a lot of fragmentation in the industrial manufacturing market, said Ryan Martin, industrial and manufacturing research director at ABI Research.
“It is difficult to speak broadly about those technologies given the diversity of manufacturers and the problems that they each face,” said Martin. “There is a huge base of small and mid-sized enterprises and machine shops, especially in the U.S., which employ one to 10 or one to 50 employees that are really critical and could benefit from these technologies but are not necessarily employing them or have great access to them.”
Martin said that only a few large companies, such as household names like John Deere, Caterpillar, GM, Tesla, Apple, and Samsung, are driving a lot of the innovation.
For these types of larger companies, they are typically employing AI in both the information-technology and operational-technology sides of their businesses — the former for data management and the latter for monitoring and control of industrial equipment.
The two main areas are data-related, which could be data normalization, data cleaning, or data analytics, so extracting insight from data and and making that data accessible, said Martin. The other area is in quality, which can be quality in products or machines, including predictive-maintenance–type applications, he added.
Machine vision is another area where AI is being deployed more widely for applications such as anomaly detection, said Martin. “What’s changing there is the difference between AI and deep-learning–type applications that are looking for anomalies that are not predetermined.”
That’s really important because historically, these systems are good at finding issues if they’re known issues, but one of the biggest challenges is finding and fixing issues that you don’t know you’re searching for, and that is where AI more broadly comes into play, he added. “Ideally, you would be able to tie that anomaly back to the reason and then initiate an action from it. Maybe it’s identifying that the defect happened because of an issue with the machine that’s producing it, or it could be that the defect occurred because of a supplier issue and that action could be taken automatically.
“Today, it’s more likely that a person or several people are involved in those processes to identify the issue and then implement its corrective action,” he added.
AI is also employed in the design phase, with AI embedded into the design software, said Martin. With generative design, designers input their key product parameters, and the software using AI ideally generates a number of designs, he added.
The designer then narrows down the design from the list of possibilities based ON their criteria.
One example metric could be for sustainability, whereby the product selection is based on the least amount of materials that are sourced locally, said Martin. “This saves the designer a tremendous amount of time and effort.”
However, industrial manufacturers still face challenges in implementing AI models. A big part of it is a knowledge barrier to entry.
“Standing up an AI model doesn’t happen overnight,” said Martin. “You need to understand all of your inputs and, more importantly, what it is that you’re trying to achieve. There’s often a lot of setup required, and even in instances where companies can get up and running in 24 to 48 hours or even a week, which may be true, you still need to have highly skilled people.”
Martin said it is very unlikely that an AI solution can be set up overnight because data needs to be collected and analyzed and algorithms have to be developed. “Even if the common scenario these days is that a provider may come with algorithms that can get you 80% of the way there, then you have to customize the last mile or last 20%, which is a great approach, but it still does require that last-mile customization, which requires time or partners.”
But there is change happening in the industry thanks to new software delivery options. Software as a service (SaaS) and cloud services make newer technologies more accessible.
“Cloud as an architecture and SaaS as a delivery mechanism means that there are much lower barriers to entry and manufacturers can get up and running in a very short amount of time and ideally without as much training because all the infrastructure would be supported and enabled by another partner or by the provider,” said Martin.
A few examples include Siemens’s recently launched Xcelerator as a Service portfolio, PTC’s evolving portfolio on its Atlas SaaS platform, and AutoDesk’s Fusion 360.
Component manufacturers are also focused on reducing the barriers to entry. One example is Sensata Technologies’ Sensata IQ platform, which makes it easier to deploy asset health monitoring to prevent unplanned downtime in manufacturing environments. This cloud-based platform uses AI to process data from Sensata sensors as well as qualified third-party sensors to monitor assets from anywhere, including a PC, smartphone, or tablet.
Sensata’s solution targets the 85% of a plant’s assets that are unmonitored today. “The majority of what has been monitored today are very critical assets in a plant, and a lot of those solutions are very costly, don’t use the cloud, and are integrated into the control systems,” said Bryan Siafakas, product line director of Sensata’s industrial sensing and IIoT portfolio. “Our focus is on that balance of plant assets where we don’t have to tie into the existing control infrastructure. It is easily retrofitable on any asset that you might want to monitor in the plant.”
The learning period to build a baseline of an asset could take one or two weeks. That information is then stored and trended in Sensata IQ, leveraging AI/ML, which is used to monitor for certain types of faults. Sensata pegs its fault characterization accuracy at 95% based on benchmarks.
Some of the anomaly detection (via the sensors) is done on the edge, and when required, data is sent to the cloud for more sophisticated analysis that requires the extra computational horsepower. It depends on what data is required to be able to characterize the fault, said Siafakas.
He said the ease of use starts with deploying Sensata’s sensors, which can be configured or installed with no tools. As an example, one of its wireless vibration sensors comes with a magnet mount, which can be placed on a motor or pump and configured with a Sensata IQ mobile app.
“Answer a few easy questions and then you can visualize it in the platform,” said Siafakas. “What else makes it easy is that AI machine learning takes the domain expertise required from the customer and packages it into models that are running in the system.
“AI allows the system to take on that domain expertise so that the maintenance manager or plant management can interpret those signals in an easy way, alerting you that there will be an anomaly or particular fault in the system,” he added.
The key point is to avoid downtime, interrupting the fault before it actually happens and brings a plant down, said Siafakas.
Sensata IQ enables factory managers and maintenance engineers to monitor all their assets from anywhere, including on a PC, smartphone, or tablet. (Source: Sensata Technologies)
These types of solutions can also help solve some of the skilled labor shortages in manufacturing plants.
One of the key trends that Industry 4.0 is looking to address is the skills gap as baby boomers retire and manufacturers struggle to replace the skilled labor, said Siafakas. One example is “maintenance personnel who have been in a plant for 30 years and can just walk by an asset and say, ‘That doesn’t sound right.’ Newer entrants into the workforce don’t have that same level of experience or depth of knowledge, and now you can sensorize those assets, leveraging AI to be able to predict those faults.”
AI is a really good fit for redundant tasks that can be automated, said Martin. “It’s striking the right balance between software-driven automation and human-driven action insight that allows people to do what it is they want to do but also have them empowered with the right information so that they are physically where they need to be when they need to be there and doing things in an efficient and optimized manner. That could include not just the manufacturing of the physical product but all the way through the design or after-sales service and support.”