GenAI in Manufacturing Production Wonders via Prompt?

Related Vendors

FAULHABER Drive Systems

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

Wuxi InfiMotion Technology Co., Ltd.

Until now, the use of artificial intelligence in manufacturing has mainly been based on machine learning: algorithms recognize patterns in historical data in order to make predictions or detect anomalies. Generative AI is now opening up completely new fields of application for the industry in terms of knowledge management and process automation. However, the technology is not a sure-fire success.

The production halls are becoming increasingly intelligent: machines that think along, robots that understand their environment, and systems that engage in lively conversations with technicians. This sounds like science fiction, but it is already possible today and will be the key in the future to aligning diverse manufacturing processes, individual customer requirements, and economic pressures.

The fact is: While "classical" AI in industrial environments is more or less an integral part of many operations, primarily in pattern recognition, quality control, and predictive maintenance, GenAI introduces a new level of capability. It independently generates design variants and simulates feasibilities. Agentic AI goes even further. The technology acts autonomously within defined target parameters—from automatically adjusting machine settings to performing complex troubleshooting processes without human intervention.

The Promise of GenAI And Co. is Convincing

The advantages promised by GenAI and similar technologies are certainly convincing. In the study "From the Factory Floor to the AI Era" by NTT DATA, 95 percent of surveyed executives see generative AI as a kind of super booster—accelerating processes and improving production outcomes. 91 percent are convinced that digital twins combined with GenAI will make both physical assets more efficient and supply chains more robust.

The reality, however, often looks different: Many manufacturers have production lines with mechanical automation solutions and industrial robots, but broadly scaled GenAI projects are rather rare. Especially when it comes to the shop floor or integrating the entire supply chain into AI-optimized processes, many manufacturers remain hesitant. This may be justified for purely rule-based machines and processes, where GenAI might be too imprecise. In all other cases, however, the industry is missing out on a real opportunity.

Why? GenAI helps make internal knowledge accessible to everyone. The technology can generate maintenance manuals in natural language, visualize step-by-step processes, or provide augmented reality applications with context-specific information. This plays a particularly important role in onboarding new skilled workers: virtual assistants that access machine or production data in real time help convey knowledge and minimize work errors.

In the past, engineers needed years to understand everything. Today, a newcomer can use an AI-based design program to, for example, find the best way to mill the desired product design from a raw material. Instead of laboriously searching for the right information, they can directly chat with manuals, maintenance reports, and other information sources. Furthermore, generative AI can develop dozens of practical design alternatives in a matter of seconds. These are automatically checked for feasibility, allowing the engineer to focus on the final selection and validation.

Another example is the creation of requirement specifications, which remains a bottleneck in machinery and plant engineering. GenAI automates large portions of this work by automatically generating specifications in line with legal requirements from historical projects, product data, and customer demands. This allows engineering capacity to be used for refinement and validation rather than investing time in repetitive documentation. Especially in the design-to-order sector, where individual adjustments are required, this significantly reduces lead times. Generative AI is essentially the perfect sparring partner for questioning and improving existing processes.

Physical AI Bridges the Gap Between the Digital and Real World

In the next development stage, the technology can be linked with digital assistants. The buzzword is “Agentic AI”: Instead of merely executing pre-programmed workflows, these systems pursue a specific goal, such as a defined product quality or minimal scrap rate, and independently adjust process parameters to changing conditions. An example: If the viscosity of a lubricant changes due to temperature fluctuations, the agent adjusts the pressure or speed in a CNC machine without any operator intervention.

In highly flexible production lines processing multiple product variants in parallel, multiple agents act in coordination. They communicate with each other about resources such as machine time or tool changes and optimize the sequence of work steps. This creates an adaptive production system that can respond to fluctuating order volumes or changed customer requirements in real time.

Physical AI finally expands generative AI models with a deep understanding of spatial relationships, material behavior, and physical laws. Simply put, it is about transferring human intuition and experience to machinery. While a skilled worker may know their own machine well, they lack this knowledge regarding the entire production line. Of course, they can predict far in advance whether something is wrong with the machine and could subsequently affect the production outcome based on suspicious vibrations. However, their perception remains limited.

AI, on the other hand, captures thousands of pieces of information in a fraction of a second. This is where Physical AI showcases its strengths. The foundation lies in precise simulations of factory halls, machines, and robots in a digital twin. In this virtual environment, the systems learn through simulations that mimic real scenarios how to navigate safely in real spaces, grasp objects, or collaborate with humans without collisions.

For manufacturing companies, this means: A collaborative robot assists with assemblies by proactively anticipating human movements. A manipulator, in turn, adjusts its gripping force to the actual weight and shape of an object.

Avoiding Pitfalls in Implementation

As promising as the potential of GenAI, Agentic AI, and Physical AI may be, their implementation poses technical and organizational challenges for manufacturing companies. First, a clean data foundation is required. Process, machine, and quality data must be standardized, accessible, and interconnected. Additionally, real-time decision-making capability requires agents deeply integrated into MES and SCADA systems.

Furthermore, it makes sense to leverage the knowledge of your own employees. An interview bot can be used to identify potential pitfalls that might not have been on anyone's radar yet. At the same time, the question of IT infrastructure arises, as foundation models for GenAI require enormous computing resources.

In practice, many manufacturers therefore combine private cloud systems for sensitive operational data with public cloud capacities for computationally intensive training runs. Real-time decisions, however, mostly take place close to the edge in the production line itself. Governance is also becoming more important: Who is responsible if an autonomous agent makes a wrong decision? What rules must apply to safety-critical processes? Manufacturers must define clear guidelines here and train agents to operate within safe parameters.

Additionally, there is the question of AI: Large language models (LLMs) like GPT, Anthropic, and Gemini dominate the market. They assist in drafting emails, creating images and videos, or programming. However, they do not understand the manufacturing language, which, in addition to texts, recordings, and code, also includes 3D models, CAD files, and sensor data with countless parameters such as humidity, circuit diagrams, industry, and environmental standards.

Therefore, task-specific optimization within the framework of SLMs is sensible. They can be fine-tuned for specific areas, making them more efficient and enabling higher accuracy. Unlike LLMs, SLMs can also run locally on edge devices. This eliminates the latency associated with cloud-based processing, allowing decisions to be made almost in real-time.

An External Partner Contributes Their Wealth of Knowledge

The latest AI technologies can definitely accelerate the innovation cycle in manufacturing significantly. However, most manufacturers lack the necessary resources. Therefore, involving an external partner is advisable in most cases. They bring the required expertise, including experience from customer projects in other industries.

This opens up an entirely new perspective. It aids in the validation of use cases and the most suitable technologies, as AI is not always automatically the best solution. With its blueprints and best-of-breed approach, it also ensures that projects achieve success quickly. The goal is the progression from assistance to partial automation and ultimately to full automation, where value creation encompasses the entire manufacturing cycle.

Machine Authentication

Machine Operator Authentication - Why knowing "who" matters