AI in manufacturing – from skepticism to mass adoption
AI was everywhere at the Automation Fair, in various forms from generative AI to machine learning to solutions such as ChatGPT with applications across the product lifecycle from product design, through to optimizing supply chains, and everywhere in-between. Over the last year the technology has gone through a shift in acceptance. According to Judson Althoff, EVP & Chief Commercial Officer at Microsoft, one year ago, people were skeptical about AI – treating it like black magic, then over the last three months, this switched to the fear of its impact on our future and now users are “coming up with hundreds of ideas” for the applications for AI and ChatGPT.
These opportunities continue to expand along with the technology. Nicholas Thompson (CEO, The Atlantic) in his keynote highlighted the technology transformation as a result of Moore’s law and then highlighted the opportunities ahead stating, “AI is more like 10 times Moore's law.”
"AI is more like 10 times Moore’s law."
“The PLC of the future will have native AI capabilities.”
It’s not just the power of AI, it’s how widely it can be deployed across different equipment. According to Cyril Perducat (CTO) the “The PLC of the future will have native AI capabilities.” These capabilities will be applied in improving the worker experience and effectiveness as well as in optimizing varying parts of the manufacturing process.
Design AI can influence every stage of design, examples shared included the redesigning of manufacturing processes so different products can be manufactured with limited changeovers, support with virtual commissioning as well as in the design of code.
Althoff noted that that 30% of the code for Copilot projects were written themselves by generative AI.
Closed-loop operations In operations, machine learning can be utilized to support closed loop optimization of manufacturing processes. This can be done by autonomously adapting to variable parameters such as ambient conditions, input material etc. to optimize the process and automatically adjust for disruptions. This transition “from automated to autonomous will enable people to focus on other tasks” according to Perducat. Another example provided by Perducat was in utilizing Large Language Models (LLMs) to understand and improve old and undocumented code – a particular challenge with the transition away from older programming languages such as ladder logic.
Maintenance At the show Rockwell also presented in FactoryTalk Analytics Guardian AI, a machine learning solution to support edge-based analytics, with the ability to monitor asset health from the data collected from Rockwell's variable frequency drives (VFD).
Also highlighted was Fiix, recently acquired by Rockwell, which delivers a standalone Computerized Maintenance Management System (CMMS) and AI-powered Asset Risk Predictor (ARP). All these solutions are designed to improve the effectiveness of maintenance teams to quickly identify and repair deteriorating assets. In the longer term, Guardian AI and Fiix may have some integrations to further deliver advanced maintenance.
"We’re adding generative AI copilots because they can augment our users, open up the platforms and ecosystems they need to reach, and let them and their partners co-innovate, using software like Rockwell Automation’s FactoryTalk Design Studio."
Supply Chain For supply chain solutions, machine learning is, according to Shepherd, already available to support better demand planning, which has led “demand forecast accuracy improved by 5% to 30%” for customers, with Plex.
In spite of all of the opportunities provided, people need to be kept in the loop. Blake Moret, CEO of Rockwell Automation said Rockwell are co-innovating with Microsoft and adding generative AI copilots to its configurations and other tools.
“We’re adding generative AI copilots because they can augment our users, open up the platforms and ecosystems they need to reach, and let them and their partners co-innovate, using software like Rockwell Automation’s FactoryTalk Design Studio,” (source Control.com). Copilots are chatbot type assistants to support workers in their decision making and counter the challenges of workforce shortages.
The manufacturing sector has traditionally been a notoriously slow-moving, technology adoption laggard, with those in the industry more reluctant to introduce new approaches and solutions. However, when it comes to confidence in the value of AI rates of acceptance are soaring. According to Omdia’s AI Maturity in Manufacturing 2023 survey, 85% of manufacturers increased their confidence in AI’s ability to add value to their operations.
The focus on AI in the manufacturing sector is being driven by pressure for efficiency improvements and cost saving. Recently added are sustainability targets, energy efficiency, as well as improved customer experience. Shortening products’ life cycles and the need for more flexibility also requires support from digital technologies. The lack of a skilled workforce and workers’ safety enablement are another manufacturing specific focus that AI can be leveraged to address.
AI-based chatbots and virtual assistants are the one of the three most adopted applications in manufacturing. These are collaborative tools that augment the workforce and make existing know-how, including programming, easier to access and to implement. Usage of the Large Language Models (LLMs) for code generation not only simplifies the code development, but also makes it more time efficient, consistent and allows fast prototyping. Microsoft is well rooted in manufacturing space and has heavily invested in AI solutions in the last year. Pairing this with the domain expertise of automation companies is a good prerequisite for the successful partnerships on co-developed AI solutions.
Although there are clear and compelling use cases for both AI and generative AI in manufacturing, there are also a number of companies that companies must deal with to ensure successful implementation, including data reliability and standardization, data governance. Challenges of scalability, complexity of integration and security are also still to be overcome.
