Artificial intelligence (AI) is here to stay and its future possibilities are endless. Advancements in connected machines and data analytics combined with AI and machine learning on the shop floor are fueling innovation and accelerating manufacturing transformation.

As the manufacturing sector embraces AI, technology leaders must understand AI’s potential impact on factory operations and workforce strategies. What are the opportunities and risks AI offers in manufacturing? How can manufacturers successfully implement AI and prepare their workforce to integrate it into their processes? What’s its future outlook? Tech Briefs asked four industry experts.

Our roundtable participants include Himanshu Iyer, Senior Industry Manager of Manufacturing at NVIDIA; Edward Mehr, CEO and Co-Founder of Machina Labs; Jim Eskew, Senior Director of Product Management, Automated Machine Learning at SparkCognition; and Dheeraj Vemula, Digital Twin - Business Development at Altair.

Tech Briefs: What are the biggest AI trends in 2024?
Himanshu Iyer

Himanshu Iyer: The fusion of industrial digitalization with generative AI, fueled by advancements like large language models (LLMs), is set to digitalize our physical realm, turning processes and products into data. This digitalized environment will allow AI to help optimize real-world processes. It will also become a fertile training field for AI itself, improving its functionality. This symbiotic relationship is set to drive unprecedented innovation and efficiency across industries. Meanwhile, retrieval-augmented generation (RAG) will bolster accuracy and reliability of generative AI. It allows a generative AI model to access real-time information not included in its training data to enhance its responses to prompts. RAG promises to advance generative AI applications in manufacturing, meaning more enterprises will look for ways to incorporate it into their production-grade AI.

Edward Mehr: Manufacturing has been a late adopter of technology, but we will start to see more integration of AI into manufacturing processes that will continue to accelerate. With AI, you can automate processes that aren’t as fixed, such as QA and defect detection, which has been well documented and adopted. We’re starting to see preventative maintenance as an area where AI can be applied now that machines are starting to connect and sensor data is collected. The next generation of AI in manufacturing is to fully integrate into the whole manufacturing process including the manipulation of materials. Another area that we will see more of is the use of LLMs to simplify the human interface in the supply chain. AI will define and gather requirements that will automate the design and communication requirements.

Jim Eskew

Jim Eskew: Three areas that will benefit in ways that have been previously out of reach include mining unstructured data, curating computer vision use cases, and edge analytics. Gartner points out that 80-90 percent of enterprise data is unstructured, i.e., documents, image, records. It’s now possible to not only query and find keywords in digital documents, but now to translate, synthesize, and cite text embedded within scanned documents and images based on natural language prompts. Similarly, we see ‘information overload’ as a big challenge. Too many monitors and not enough personnel, along with fuzzy images make utilization of video feeds less effective. Improvements from graphics cards together with generative techniques to improve image quality, combined with platform technology that enables the build-out of more specified vision use cases make for scenarios where operators can leverage their cameras to do more with less. Not to be overlooked is the continued advancement in communication technology, including LEO satellite clusters along with highly networked facilities. Because edge computing and data transmission are now so much more readily available and affordable, plants and distributed assets will become much more connected and able to relay warning signs of impending failure, giving ample time to prepare and respond.

Dheeraj Vemula: As seen over the last year, AI/machine learning and data analytics are playing an increasing role in how businesses view market trends, ground their predictions, and make decisions. With the latest advancements in predictive analytics tools, including no-code/low-code platforms, the analysis and interpretation of data is becoming easier and quicker. This opens up enormous opportunities in the product design cycle as engineering teams realize the benefits of combining data-driven approaches with physics-based simulations. The proliferation of LLMs and generative AI also unveils opportunities for unparalleled productivity spanning from the creation and execution of innovative ideas to offering highly personalized experiences. MLOps (machine learning operations) is set to undergo significant advancement, expanding beyond operational functionalities like deployment, scaling, monitoring, etc., to encompass model optimization for more edge devices, equipment, and appliances. However, with the recent executive order on AI from the White House, users will demand a more transparent understanding of their AI journey with “Explainable AI” and a way to show that all steps meet governance and compliance regulations.

Tech Briefs: What are the opportunities and risks AI offers in manufacturing?

Himanshu Iyer: Digital twins — supported by AI and learning models — allow manufacturing processes to be simulated in increasingly complex variations, from inception to physical replication. This reduces risks to end users, supports collaborative work, and enables formerly sequential steps to be parallelized. Data interoperability and creation platforms that can access these streams are key requirements. AI can now be used to build learning models and synthesize data, improving existing state models and helping create new ones. An AI-enhanced digital twin can simulate scenarios and suggest alternatives based on optimization parameters set by operators. The processing power required may grow at increasingly higher rates, which is where optimized data centers come in. Optimizing power consumption and processing factors will help mitigate carbon footprints. These data centers, of course, will harness their own digital twins and complex simulations to achieve these goals.

Edward Mehr: From our end, the big opportunity for AI is to move away from traditional repeatable automation to custom and situationally aware automation. AI brings that flexibility so that you don’t have to create machines that do just one thing, but that you can have machines that build multiple parts independent of design and material. One of the challenges we see in AI in manufacturing is how this new way of manipulation that is so flexible will measure up on the qualification side. We’re used to systems doing the same things over and over again. With AI, we will have repeatability but since the system is dynamic, we need to prove that it’s still repeatable and could adhere to the same quality specifications.

Jim Eskew: Hyper-personalized solutions that curate the dissemination of information will unlock value for all range of users. Imagine the learning curve for a new hire if all the institutional (i.e. ‘tribal’) knowledge that experienced employees possess are made available on demand in an accessible way, especially if they can specify how verbose that responses from such a search are and engage in a dialogue with the material. This will translate to higher asset availability and safer work environments, even with turnover in staff. The risks posed by AI is precisely the nature of the models themselves, they are stochastic in nature meaning all analysis is still approximate. This means there will be false positives and LLM hallucinations, even with safeguards added to the models. To protect against inaccurate information leading to false conclusions, AI systems should present evidence — for example if an anomaly is detected, what the factors that led to this anomaly, how does it compare to previous anomalies, what reference points are available; for LLMs, what is the source documentation that led to the response, etc. AI models can supplement expert system, but need to provide transparency to ensure correct, actionable conclusions.

Dheeraj Vemula

Dheeraj Vemula: An increasing number of manufacturing firms have started adopting AI to optimize their processes and improve overall efficiency. One of the more exciting developments gaining momentum is the integration of data/AI with simulation in the design process. By applying data-driven insights to a virtual model, commonly known as a digital twin, design engineers can quickly run thousands of iterations and test different product scenarios before going to prototype. This approach leads to accelerated time to market, reduced costs, and minimized material usage, which contributes to manufacturers’ sustainability goals. Moreover, insights from data gathered in previous projects can be leveraged for new designs, preventing valuable knowledge from being overlooked. While more manufacturers are embracing AI and data analytics strategies, they are also experiencing considerable challenges. These include workforce issues related to their inability to process data quickly, inability to use data to make informed decisions, and issues with data quality.

Tech Briefs: Despite this opportunity, many executives remain unsure where to apply AI solutions. What advice do you have for manufacturers who want to implement AI in their design and production processes?

Himanshu Iyer: The first wave of AI is about digital intelligence running in data centers. The next wave of AI will be about software increasingly coming out of computers and data centers and running in the world, bringing automation to sites such as plants, buildings, and factories. To develop such software, virtual worlds must be built where AI models can learn and be tested before production. This process is key for creating physically based, accurate digital twins of real-world industrial systems. Companies are increasingly using digital methods to design, test, and configure physical products for optimal performance. AI-enabled digital twins can bring unprecedented acceleration and novel capabilities to businesses by allowing them to accurately predict and optimize their physical products. Ultimately, the digitalization of complex industrial processes with the help of AI will be used at scale across industries such as healthcare, manufacturing, automotive, semiconductors, and more.

Edward Mehr

Edward Mehr: AI in and out of itself is just a technique, a tool. What we see that works for our customers is to identify the business case first. In many cases, you need to partner with people and experts to solve a specific problem. Direct implementation of AI is too abstract, but solving a problem using AI is something that business executives can get behind. Some of the solutions might involve AI, but there could be other ways to solve the business problem.

Jim Eskew: Simply put — get started. At this stage in the game, if you are not adopting AI, you are falling behind. Competitors who leverage AI will have higher production rates, with less supply chain disruptions, lower overall cost, and lower defect rates. AI really can and is impacting this wide array of value drivers in manufacturing. As a corollary, because there is an abundance of opportunity, there are many ways to go about achieving results. Some ways to see results are: 1) Set up and run proofs of concept with vendors or with your own staff leveraging open source or cloud software; 2) Purchase solutions that address specific use cases such as logistics planning; 3) Try out platform technology that enables the rapid deployment of AI solutions for specific classes of models e.g., computer vision, natural language processing, anomaly detection, etc. AI adoption is a journey, but any way you go about it will undoubtedly yield value and prepare you to target more areas of improvement.

Dheeraj Vemula: The exact solution depends on the specific manufacturer’s use case since AI solutions do not come one-size-fits-all. At the same time, there are certain key criteria that one can look for when trying to apply AI. The best practice is to start with the business challenge that affects the bottom line most and work backward from there. The second criteria is the availability of data. For the AI solution to be effective there needs to be valuable data, the kind of data that can allow for meaningful insights. Once a manufacturer can identify a sweet spot of use cases, where there is a genuine business problem and data available, an AI solution can be tailored to meet the specific needs of the manufacturer. For example, imagine a continuous process where the smallest variation in the manufacturing process will affect the entire output batch and leads to discarding all the output. If we have data about the process parameters (through on-board sensors), AI can help identify any anomalies in the process parameters and enable the operator to take corrective action before wasting a lot of material.

Tech Briefs: The accelerated adoption of AI in manufacturing raises concerns about job displacement and there is resistance by employees to adopt AI tools. How can manufacturers prepare their workforce to integrate AI into their processes?

Himanshu Iyer: AI isn’t a job killer; it’s a job changer. Plus, it has the potential to create new jobs that we can’t even imagine today. In the future, manufacturing facilities will need to be digital-first to stay competitive. Manufacturers today are experiencing external pressures such as periods of high inflation, rising energy prices, and a pressing need for greater sustainability. To address these challenges, companies of all sizes are increasingly investing in digitalization, simulation, and decarbonization. Generative AI has also ushered in a new era of advanced language and design workflows. For example, machine operators can interact with a copilot AI in natural language to identify problems and opportunities for improvement. Generative AI also enables designers to craft better products by accelerating the ideation process, providing review options, and ensuring company branding alignment.By tapping industrial digitalization applications early, manufacturers can be the first to develop new services and unlock operational efficiencies across production lines. Humanity has no shortage of great ideas. AI will help more people tackle more of them.

Edward Mehr: There is a perceived risk of job loss with AI, but at the moment we have a worker deficit. The industry simply has a hard time finding people to do the jobs. We see automation as a way to bring the employees in, and in this case, we’ve never had a situation where people are resistant to the use of new technology or AI. Also, new technology systems require new skills. We’re adding jobs that traditionally weren’t in manufacturing but rather in the IT industry. To attract that skill set to the manufacturing industry, we need to develop a different culture that manufacturing is an enticing place to be.

Jim Eskew: In any manufacturing environment, the challenges that AI targets are primarily equipment reliability, quality control, and safety. This does not take away from the need to have operators and maintainers, but rather the other way around — workers who adopt the technology will be associated with improved KPIs. The key to adoption ultimately is easy access and intuitive applications for these users. Many applications of AI can be leveraged with laptops or even mobile devices. For instance, visual inspection applications leveraging computer vision AI algorithms can be used with a cell phone, comparing produced parts to design specs similar to processing a check on a mobile banking app.

Dheeraj Vemula: As AI continues to gain traction in the manufacturing industry, there is a growing demand for additional data scientists to effectively analyze and apply the data to ensure the benefits initially intended are achieved. No-code or low-code solutions make it possible for people without much programming experience to build and deploy data science models. This can help to democratize data science and make it more accessible to everyone, including knowledge experts in a specific domain (i.e., design engineers). These no-code solutions empower users to work with data analytics models without ever writing a single line of code. Additionally, these tools enable data scientists to create reusable solutions that can be applied to a wide range of use cases, saving time and effort as well as improving the quality of data science projects. As for concerns and uncertainties related to AI and job displacement, certain aspects of work, particularly in manufacturing, will get more automated. But to facilitate AI adoption, you still need “lots” of people, whose work will be enhanced as they spend more time on the less time-consuming tasks and focus more on the strategic aspects of their role.

Tech Briefs: How will AI progress over the next five years? In which industries will we see the biggest adoption?

Himanshu Iyer: Companies of all sizes will move to address today’s challenges and future ones by increasingly investing in and embracing the possibilities of industrial digitalization and decarbonization. We’re confident that every factory will be digital-first in order to stay competitive. By leveraging industrial digitalization applications such as advanced simulation and digital twins supported by AI, manufacturers will be able to develop new products and services and unlock speed, agility, and operational efficiencies across their enterprises and value chains.

Edward Mehr: Traditionally, aerospace has been an early adopter of emerging technology so it will continue to adopt AI. What will be new are the horizontal applications of AI in industries that will open up, especially where configure to order and customization is key. We are starting to see that in HVAC and automotive.

Jim Eskew: AI will progress in two key ways: First will be a growing comfort with AI assistants — powered by LLMs — for virtually all aspects of life. In particular, multimodality will be key to unlocking the benefits of generative AI. In other words, where there are many different sources of data, those sources are distributed without logical relationships between them, and there is a compelling need to retrieve information in context quickly, AI will supercharge the experience. Where this comes in handy is for operators of large fleets, i.e. of aircraft, gas turbines, or vehicles, where this is a high volume of technical malfunctions, sensors tracking key measurements, a deep corpus of documentation describing incidents, and a variety of parts required to make repairs. All of this information can be compared in one spot, on the fly, and saved for reference. The second key way that AI will augment industry is having ‘eyes on’ distributed assets with the assistance of computer vision algorithms. The main benefactor will be those looking to improve safety and security; this can be workplace incidences from hazards, such as suspended loads, or it could be schools or facilities looking to protect against would-be attacks.

Dheeraj Vemula: The growth of AI has been exponential, and the technology has matured very well already. Adoption is now catching up. Over the next five years, we will see AI being adopted everywhere. Right now, AI is being applied by only experts and data scientists, but with the growth of low-code and no-code solutions, we will see AI being democratized to everyone within an organization no matter their role. The biggest benefit of AI is improving productivity. Over the next five years, we will see AI adopted the most where productivity matters the greatest. Especially industries like manufacturing and operations, because of the data that is available due to Industry 4.0. We are also seeing early adoption of AI in healthcare — drug manufacturing, diagnostics with imaging etc. That is also likely to accelerate given the amount of cost savings AI can provide.

For more information, visit: