Generative AI in Manufacturing: an Integrator’s Perspective

• General,
• Manufacturing,

| December 11, 2024

The emergence of generative AI is rapidly reaching its tendrils into many facets of society and its infrastructure, which is generating about as much excitement as it is fear. Educational organizations are concerned with plagiarism; the art world is concerned with data collection of original works; government and social organizations are concerned with disinformation and misinformation.

Perhaps the manufacturing industry is one of the few that is able to make the most of generative AI without the other side of that double-edged sword. Where other industries are struggling to reign in this unbridled new technology space, manufacturers are looking to adopt it and apply it to as much as they can. But that comes with its own set of struggles.

E Tech Group, a leader in automation and engineering services, is at the forefront of these developments, helping clients stay at the forefront of their sector. In this article, we discuss the potentials and pitfalls of industrial applications for generative AI.

Understanding the Capabilities of Generative AI

Generative AI creates new content, whether text, image, music, etc. by mimicking the human learning process. Where traditional AI is often based in recognizing patterns and making predictions based on available data, generative AI is meant to produce novel outputs. Here’s a breakdown of how it works:

• Training Data: Generative AI models are trained on large datasets that contain examples of the type of content they are intended to generate. For instance, a model trained on thousands of paintings might learn to create new artwork in a similar style.
• Learning Patterns: During training, the AI learns patterns and structures from the data. For text, it might learn grammar, vocabulary and context. For images, it learns shapes, colors and composition.
• Generation Process: Once trained, the model can generate new content by using the patterns it has learned. For instance, it can write an essay, create a new piece of art, generate chat responses, designs, and even create synthetic data based off previous inputs.
• Applications: Generative AI is used in various fields, including content creation (like generating marketing copy or news articles), art and design (creating visual art or graphics), and even scientific research (designing new molecules or compounds).

Generative AI leverages learning techniques such as neural networks, particularly Generative Adversarial Networks (GANs) and transformer models, to produce results that are often indistinguishable from their manmade counterparts.

 But how does all of this apply to manufacturing and industry?

In current industrial applications, Generative AI (GenAI) leverages advanced language models like ChatGPT, Gemini, and open-source models like LLaMA.

However, these large language models can sometimes generate inaccurate or irrelevant responses, known as ‘hallucinations.’

To mitigate this, a technique called Retrieval-Augmented Generation (RAG) is employed. RAG enhances the existing LLM models by providing them with access to relevant information, such as manufacturing data, beyond their original training data. This additional context helps the models generate more accurate and informative responses, ultimately improving the effectiveness of GenAI applications.

> 

Use Cases for Gen AI in Process Automation

Because of its capabilities, generative AI has the potential to transform automation in manufacturing. It’s quickly reshaping what the digital transformation looks like, as it can enhance every step of the value chain:

Intelligent Product Quality Control

Just like an employee inspecting products for defects or controlling the parameters of an industrial mixer, generative AI can be trained to look for signs of quality control problems. They can create diverse datasets of product information – images, specifications, defects, etc – to train a robotics vision system to do automated inspections.

But gen AI can go a step further by utilizing the data it’s received and generated to simulate various manufacturing conditions. This offers the users several quality control benefits:

• A more dynamic and flexible approach to identifying issues. Unlike traditional systems, which rely on predefined rules and parameters, generative AI draws on vast datasets and its ability to learn patterns, enabling it to detect anomalies that weren’t explicitly programmed. For instance, AI could identify subtle defect patterns across product lines that might go unnoticed by humans or traditional automation systems.
• Improved ability to correctly detect and identify defects more accurately and precisely than human capabilities.
• Training and validating quality control systems in a digital environment before on-the-floor implementation, also known as digital twin.

With AI integration, tools like encoders, including Variational Autoencoders (VAEs), can support anomaly detection by identifying subtle deviations in product quality. This capability helps manufacturers analyze how slight changes in production impact quality and performance. Essentially, AI enhances the ability to continuously inspect, detect, and address issues in the automated manufacturing line with greater precision and minimal human intervention—delivering capabilities far beyond traditional methods.

Improved Predictive Maintenance

Even the best system integrators can’t make a legacy platform or vintage equipment live forever, and every manufacturer knows the operational and financial costs associated with poor organization/understanding of when control systems, their hardware, or automation equipment needs maintenance or replaced. Decreased throughput and product quality, increased errors and downtime, bottlenecks to adapting and scaling – these can be make-or-break factors.

The ability of generative AI to inspect, gather operational data, analyze it, and learn from it allows it to create predictive maintenance plans. This alleviates several of the difficulties associated with the waiting-for-the-other-shoe-to-drop method many manufacturers have resigned themselves to: 

1. Precise predictions of equipment wear and age allow companies to plan for big expenses such as control system upgrades or equipment replacement.
2. Understanding the expenses associated with maintenance, planning improves budgetary practices, but also streamlines the downtime necessary to implement improvements.
3. Gen AI’s massive collection of operational data better informs decision-makers on which products and platforms are best suited for said improvements.

Consider how this then plays back to support product quality control, and we begin to see the multifaceted advantages of having a generative AI – it’s a facility-wide assistant!

Enhanced Utilization of Resources

In that same vein of thought, if we know that gen AI can (1) collect, analyze and learn from data, (2) make decisions and recommendations based on previous patterns and (3) simulate data and environments to predict the best way forward, then it only follows we will also see improvements in resource processes like:

• Optimized inventory practices: The AI’s understanding of operations and patterns can better predict when new product is needed and more accurately determine how much of said resource is necessary.
• Better supply chain management: The predictive abilities of generative AI support all types of planning, including for resources like raw materials. As any company knows, supply chain disruption didn’t stop with COVID, and properly navigating those obstacles is key to avoiding shortages and downtime.
• Improved sustainability efforts: Reducing waste is crucial to your bottom line. But in these times, all manufacturers need to have their eye on sustainability as well, and factors like energy use and defect-generated waste are two pressure points in that venture. Because your generative AI has already optimized your processes and planned for the future, your operations will consequently become more sustainable.

Data-Driven Decision Making

From the shop floor to the top floor: the adoption of generative AI provides countless tools, some of which we’ve discussed in detail, but some of which there’s simply not enough time to unpack in a single article! But we don’t need to know every single thing a gen AI can do to understand how decision-making becomes more educated and effective with its help. Consider:

On the Floor:

We have discussed how operators can benefit from gen AI’s ability to suggest (or even perform) remediation, identify issues and plan for future maintenance. Understanding what’s happening and what’s to come helps operators make crucial decisions.

In the Office:

The data and analyses performed by generative AI can improve practices around employee and customer management. Chatbots can offer off-hours support for customers or streamline internal HR practices. AI can do things like look at employee leave requests, analyze staffing to determine the best way to organize your human capital and automatically update schedules.

At the Top:

For the decision-makers and investors, the culmination of investing in generative AI results in:

Strategic planning

AI can analyze vast amounts of data on financial reports, competitive analysis, market analysis, etc, to better plan for crucial moments like automation system upgrades, equipment replacement, company growth, etc.

Financial forecasting

AI models can predict future revenue and expenses by analyzing historical data, market conditions and economic indicators. This alleviates the mystery of the impact of potential investments. Gen AI can also suggest budget adjustments to optimize financial performance.

Risk management

Generative AI evaluates various risk factors by analyzing data related to market fluctuations, regulatory changes and operational vulnerabilities. For example, it might assess the financial health of suppliers and predict potential risks to the supply chain, allowing the company to develop contingency plans.

Generative AI is a Comprehensive Asset

Companies will see further benefits from generative AI here: better product planning and development increases competitiveness; understanding supply chains, demand, use, etc. streamlines inventory scheduling and budgeting; automated intelligent customer support reduces the need for unskilled labor; and improved documentation synthesis, storage and retrieval has countless beneficial implications – from compliance to scaling to sustainability.

Obstacles to the Adoption of Gen AI in Manufacturing

As with all other stages of automation, manufacturing is slow to adopt AI. Much of this has to do with cost and accessibility. But there are also other difficulties still endemic to integrating generative AI into a facility’s automation system:

Integration with Existing Control Systems

This obstacle has just as much to do with the technology as the logistics. On the tech side, integrating AI into legacy control systems can be difficult; many are not yet designed to incorporate this kind of technology. A company who has a well-designed, established automation system may find that the addition of AI will be more a bother than a benefit.

As well, the idea of adopting generative AI is off-putting for many would-be operators. The mystery of an intelligent algorithm is enough to make people nervous. And as with any retrofit to process automation in manufacturing, the training and transition can be difficult.

Bottlenecks in Trained Support & Operators

Despite how the media has catastrophized the potential of AI into a post-apocalyptic, dystopian character, generative models are only as smart as the people who write them, and only as useful as the people who understand how to best utilize them. The fact of the matter is that at this time, there aren’t a lot of engineers who can do this; the few who can are uber-talented pioneers.

The industrial sector needs time for that knowledge base to be able to trickle down and democratize across the automation and engineering industry before AI can be taken best advantage of. Luckily, E Tech Group stays ahead of the curve, which helps our clients stay at the forefront of their sectors, too. As a Main Automation Partner, our long-term relationships with clients allow us to help them smoothly transition into big changes such as these.

Limited Learning Materials for AI Models

Generative AI needs data – its entire knowledge base begins with whatever accuracy and extent of data it’s fed. Because it is emerging, just like we don’t have enough human talent to develop effective generative AI for industrial entities on a massive (and affordable) scale, we don’t have enough learning materials for AI to accomplish these either. Training an AI takes expertise, but it also takes vast – and we do mean vast – amounts of data and learning models to use. This is another obstacle that will only be remedied with time.

Data Considerations: Bias, Security & Compliance

Another thing about data – it’s only as objective as the people or programs who collected it. As such, there are a host of factors that can corrupt even the most well-written model. The potential for biases abounds. Consider:

• Historical bias: If the AI is trained on processes that are inefficient and/or outdated, it will make its predictions based on that inefficient data.
• Selection bias: If the training data for an AI model used for quality control comes predominantly from one process or product, it may perform poorly when applied to different lines or products.
• Amplification bias: A generative AI might amplify existing inefficiencies if those inefficiencies are present in the training data.
• Overfitting bias: If an AI is trained too closely on one specific process/product, it may have a hard time adapting to/accomplishing diversified tasks.
• Stereotyping bias:.If a machine learning model is trained on historical maintenance data that primarily includes data from a specific manufacturer (e.g., Vendor  A), this model may develop a bias towards that manufacturer. This could lead to under or overestimating the lifespan of other manufacturers’ equipment, the model may not accurately predict the lifespan of machines from other manufacturers, leading to unexpected breakdowns and increased downtime.
•  
• Misalignment with operational standards: If a generative AI system generates recommendations based on biased historical data or algorithms that do not align with current operational standards, it could suggest strategies that are not feasible or optimal in the present context.

We can clearly see here the theme of generative AI: it’s only as good as its training data. Limited learning material and limited data quality all hamper models and their potential applications.

On top of all that, we also have to consider security and privacy when implementing AI. AI systems learn from data – where is the data coming from? Where is the data your AI system is putting out? Is it secure? How do you know? Does the system comply with regulatory standards? Does it understand those parameters and how to stay within them while still performing its tasks? Gen AI needs direct human oversight to ensure these factors are properly handled.

Future Perspectives on GenAI in Manufacturing

Worth nearly $11 billion in 2023, the generative AI market is projected to exceed $165 billion by 2033. This growth highlights generative AI as the new frontier of industry. Beyond manufacturing, industries such as distribution, life sciences, mission-critical systems, and food and beverage are poised to benefit from and be transformed by GenAI’s capabilities. However, the success of this transformation depends on how the adoption journey is navigated.

For manufacturers, the key to achieving the best ROI lies in partnering with the right automation system integrator. Implementing advanced control systems that incorporate generative AI is not an overnight process—and it shouldn’t be. Gradual, progressive adoption minimizes disruptions while maximizing the potential of this transformative technology.

E Tech Group is uniquely positioned to guide companies through this evolution. By focusing on control system upgrades that are flexible, adaptable, and scalable, and coupling that with a collaborative approach to training and integration, we help our clients leverage generative AI effectively.

Take, for example, the growing utility of GenAI for programmers. Tools like Microsoft CoPilot are already proving valuable in understanding and generating PLC code, providing support that boosts engineering productivity. Rockwell Automation’s incorporation of generative AI into Design Studio further demonstrates how this technology is reshaping automation development. As these tools mature, their impact on efficiency and innovation will only grow.

In addition to programming, GenAI presents excellent opportunities in supply chain and asset management. Platforms like Datch are using generative AI to enhance asset management and streamline supply chain operations, delivering practical benefits today while setting the stage for further advancements.

That said, generative AI is still in its early stages for plant floor automation, where traditional AI and machine learning (ML) remain the dominant forces. While GenAI offers exciting possibilities, its integration into this space will require time, refinement, and continued innovation to realize its full potential.

By combining the right tools, technologies, and partnerships, manufacturers can unlock the transformative power of generative AI. With E Tech Group as a partner, you’ll be equipped to lead this revolution—starting with immediate gains and preparing for what’s next in industrial automation.