How LLMs are finding use in industrial applications

Feature articles | January 9, 2026

By Jean-Pierre Joosting

Industrial markets are increasingly adopting Large Language Models (LLMs) to boost productivity and streamline operations. Mostly, LLMs act as conversational ‘copilots’ that enable users to access various datasets and systems, while providing expert explanations—bridging the communication gap between people and machines. They also allow non-technical users to access advanced analytics. However, the role of LLMs in industry is expanding rapidly.

Some typical use cases include predictive maintenance, supply chain management, quality control, training, automation, and technical support.

The role of LLMs in predictive maintenance

Instead of relying on maintenance schedules or visual inspection, which can lead to over-servicing or breakdowns, manufacturers can use machine learning to ‘predict’ when to carry out a service task by identifying and comparing patterns in sensor data, performance metrics, and historical data. LLMs can be trained to analyse these patterns in real-time and interact with service personnel.

Predictive maintenance uses a variety of methods to keep machinery running. For example, vibration analysis enables manufacturing plants to reduce downtime by detecting mechanical faults early. Other techniques, such as infrared thermography to prevent overheating failures, oil analysis for early detection of wear, contamination, and degradation, acoustic monitoring to spot structural issues and ultrasonic testing to find leaks, have similarly contributed to cost savings and increased equipment lifespan in various industries.

In effect, LLMs collect and process large amounts of data from a plethora of sensors and maintenance logs. LLMs can identify anomalies and patterns in data, thereby detecting subtle changes from normal operating conditions that could indicate imminent equipment failure. Parameters such as temperature, vibration, pressure, and fluid levels are monitored in real-time. Maintenance actions or preventative measures can then be suggested to service personnel to address the situation, as LLMs can replicate the nuanced reasoning of human technicians and ‘explain’ the problem. These models are more capable than conventional analytical approaches, which passively process data, as they can synthesise quantitative sensor data within a contextual framework based on operational expertise and interact with technicians like a chatbot.

Cutting production line downtime is a key benefit. The International Society of Automation claims that factories can typically lose between 5 and 10 percent of their productive capacity due to equipment malfunctions, failures and other anomalies [1]. Downtime costs are not only due to decreased production. Other factors, such as increased scrap rates, temporary fixes, and reliance on third parties to maintain output, all add to the cost.

Retrieval-Augmented Generation (RAG) models add a retrieval step into the LLM process, enabling them to search knowledge bases and historical data related to a specific anomaly. RAGs allow technicians to make more informed decisions and receive more accurate recommendations. Using RAGs with LLMs yields more accurate predictions of imminent equipment failure, enabling technicians to prevent issues from arising and take prompt action. Predictive maintenance reduces production line downtime and cuts costs associated with unexpected equipment breakdowns. There are not enough technicians or time to carry out data-driven early detection of equipment failure or malfunction without automation.

However, challenges in using LLMs for predictive maintenance include optimising models for this specific task, ensuring data quality and addressing bias.

Managing complex supply chains

LLMs are positioned to address supply chain management (SCM) complexity by making planning more predictive. They also enable a degree of autonomous execution for inventory management and demand forecasting.

Using historical data and insights from market forecasting models, LLMs can help manufacturers predict supply chain disruptions, such as supply shortages or tariff-related cost increases. They can also respond rapidly to shifts in product demand and ensure that supply chains are optimised for specific regions. Queries and sophisticated what-if questions can be posed in natural language, resulting in informative answers such as safe stock levels, reorder adjustments, and alternative sourcing.

For example, by monitoring and analysing news and financial reports, LLMs can proactively detect potential supply chain disruptions, such as impacts on suppliers from emerging geopolitical risks or economic shifts.

To optimise logistics, a trend amongst manufacturers and logistics companies is to use LLMs as “copilots” alongside existing planning and execution systems, such as forecasting models and ERP data.

LLMs can suggest optimal routes that take into account traffic, weather, and delivery times, as well as provide predictive maintenance for fleet vehicles. Further, by analysing market conditions and historical data, they can suggest dynamic pricing models for shipping and logistics services.

As supply chains generate large amounts of documentation, LLMs are being used to automate document processing, including extracting unstructured data from documents, performing compliance checks, and generating reports.

At the customer end, LLMs have been used as chatbots and virtual assistants to provide 24-hour customer service. Since LLMs are ideal for processing unstructured data, they can serve as a unified layer across warehouse and transport management systems and supplier communications.

Unilever’s Horizon3 Labs uses GenAI/LLMs to coordinate its global supply chain operations. The planning system consolidates diverse data sources such as sales figures, meteorological forecasts, and regional market indicators and uses LLMs to provide explanations and propose adaptive adjustments for country-level planners. Reported outcomes indicate that this deployment has reduced supply chain planning errors by approximately 30%. Similarly, Maersk uses these technologies to create automated supplier contracts and obtain vendor ratings and feedback. Automation has reduced the time required for manual contract review by almost 50%. [2]

LLMs in industrial automation

In industrial automation and control, LLMs are currently used for reasoning and interaction. Currently, they occupy the layer above the PLCs, SCADA, Distributed Control System (DCS), Enterprise Resource Planning (ERP) software and Manufacturing Execution System (MES).

Today, PLCs, with their associated safety systems, form a deterministic control loop. LLMs can then be used on top of this to orchestrate, plan, interpret, and provide natural language explanations. In some cases, they may be used to generate control logic or commands that are then validated before deployment. In general, LLM outputs either inform operators or are passed through supervisory checks before modifying process variables. [3]

In control rooms, LLMs enhance human-machine interfaces by acting as natural-language control assistants, enabling operators to query plant state and operations and receive detailed explanations of problem root causes, backed by live SCADA data, alarms, maintenance history, and procedures. They can also provide voice and chat supervision of operations.

In a recent paper, researchers at Pennsylvania State University, the Institute of Automation and Information Systems, and the Technical University of Munich described how LLM-enabled multi-agent systems can enhance manufacturing flexibility by translating product specifications, CAD data, and orders into detailed process plans and resource allocations using natural language processing. The researchers describe product agents that leverage LLM capabilities to interpret evolving G-code specifications and manufacturing instructions. These agents then dynamically assign operations to appropriate resource agents, which execute tasks informed by instructions and tool specifications provided by the user. G-code is a CNC machine programming language. The paper found that GPT-4 achieved 100% success rates for two-step and 86% for four-step manufacturing processes across 50 trials for each. However, as task complexity increases, performance degrades. Next steps are to improve model accuracy and develop communication protocols incorporating an LLM for agents that can accommodate a wide variety of manufacturing instructions and requirements. [3]

Advances in using LLMs and integrating them into industrial automation, along with their use in supply chain management and predictive maintenance, are set to enable fully automated factories that can be reconfigured on the fly to manufacture a variety of products in response to resource availability, supply and demand considerations and dynamic pricing. These factories will be highly efficient and require much less human intervention.

Field servicing

LLMs are being used as intelligent co-pilots during on-site and field maintenance and troubleshooting. By integrating real-time access to service and diagnostic records, as well as technical manuals, technicians can instantly generate step-by-step guidance tailored to specific systems. The result is quicker identification of the cause of a problem and fewer repeat visits, significantly improving first-time fix rates.

LLM-driven tools are also being used to help bridge skill gaps between retiring and newer employees by providing context-based service instructions and highlighting best practices learned from past work orders. As a result, organisations achieve more efficient support operations, reduced downtime, and shorter technician training cycles while preserving and leveraging the company’s collective technical knowledge base.

Quality control

In quality control, LLMs can analyse vast streams of production data to detect anomalies and deviations that signal potential defects. The use of real-time production line data, such as temperature, pressure, vibration, vision and so on, enables LLMs to find patterns that human inspection typically misses. Deviations are flagged almost instantly—enabling rapid corrective action before defective products reach customers.

Using information from maintenance and quality control logs and reports, LLMs can recommend adjustments to machine settings, cycle times, and process variables to ensure consistent product quality. An added benefit is reduced scrap and rework costs. Companies deploying LLM-powered quality systems have achieved inspection capabilities in sub-200-millisecond timeframes, minimising error propagation throughout production runs. [4]

Further, by addressing the root causes of failures and quality defects, including material variance and process drift, LLMs are ideally placed to optimise workflows and prevent recurrent defects—resulting in lower material waste and reduced inventory requirements.

Knowledge management and training

Manufacturers are leveraging LLMs to convert static technical documentation into interactive, conversational knowledge systems that accelerate decision support and worker development. Further, RAG technology enables LLMs to instantly retrieve unstructured data such as relevant passages from technical manuals, regulatory documents, and best-practice archives, allowing technicians to use natural language queries rather than searching through dense PDFs.

LLMs capture, structure, and transfer institutional expertise to the next generation, significantly reducing training time. Manufacturers have found that LLMs offer significantly reduced costs of up to 70% versus manual documentation. Regulatory compliance is also more efficient as LLMs can interpret complex standards and ensure adherence across global operations. Organisations using knowledge-graph-powered LLM systems achieved 85% improvements in information retrieval accuracy and 60% faster certification readiness. [5]

Going forward LLMs will find wide usage

LLMs have already transformed industry and manufacturing. In the future, the models will become more targeted and accurate as well as more integrated into software, control, logistics and management systems.

LLMs will drive the development of autonomous manufacturing environments, where systems can self-configure, self-optimise, and self-diagnose with minimal human intervention. By integrating digital twins and simulation models, immediate feedback, rapid troubleshooting, and continuous process improvement are possible with real-time, unstructured data.

In CAD/CAM, LLMs will translate design concepts into manufacturing instructions and optimise performance, fully realising the design-for-manufacturing trend—aiding design exploration, reducing errors, and accelerating innovation.

Industrial automation will increasingly use LLM-powered chat and voice interfaces to manage real-time operations. With models running locally at the edge, these interfaces will enable managers and engineers to check system status, query forecasts, and resolve maintenance issues conversationally.

Enabling advanced planning and control, future LLMs will take on complex process planning, risk analysis, and production scheduling based on historical data, scenario modelling, and predictive analytics.

Future applications should also expect a push for ethical, transparent LLM development. Future models will prioritise algorithmic fairness, bias mitigation, interpretability, and responsible AI.

LLMs used to code can introduce serious errors
Edge AI SDK leverages LLMs to reduce reliance on GPUs