Currently, large language models have shown the potential impact on human-computer interaction, and the impact is staggering. The Boston Consulting Group believes that there is no doubt that generative AI will lead to a significant increase in productivity through human-machine collaboration, and industrial automation is precisely the field where human-machine interaction is most frequent.

Large language models are deep learning algorithms based on neural networks and trained by massive text data. In recent years, large language models have begun to emerge in all walks of life, especially in the field of industrial automation, the scope of application continues to expand. These models can understand and generate human language, which in turn can help greatly improve the efficiency of industrial processes. It is conservatively estimated that the large language model can help original equipment manufacturers (Oems) save 20% of their effort in building machine PLC applications.

The development of large language model is very rapid, from GPT-2 to GPT-4, the model continues to upgrade and mature, the processing power is greatly improved, and its application scope in the field of industrial automation has also been greatly expanded. As the technology matures and becomes more widely used, we expect to see more practical examples of applying large language models to industrial automation.

One of the most important features of a large language model is that it can be customized to a specific use case or task related to industrial automation. In industry, there are three possibilities for these use cases or tasks: code generation, documentation, refactoring, and testing; Natural language interface; Automation system design and development.

However, we also need to consider the limitations that may be prevalent in large language models, and how to deal with these limitations before large language models are industrialized. Next, we explore these opportunities, limitations, and how to industrialize large language models.

1. Code generation, documentation, refactoring and testing

Large language models can generate code for industrial control systems such as PLCS, or allow humans to generate human-computer interaction (HMI) pages using natural language input, simplifying the engineering process for applications and reducing the time and effort required to develop control applications. In addition, large language models are expected to improve the quality and consistency of the generated code, resulting in fewer errors and shorter debugging times. It is conservatively estimated that about 70% of the code generated by the large language model is usable for writing machine-type PLC programs, and the programming efficiency is greatly improved. Another application of the large language model is the automatic generation of recipe codes, which can save time when changing parameters, changing suppliers, or changing ingredients. The time it takes to create a recipe and trigger a line change directly affects production time, so any possible time savings help improve efficiency. Going a step further, the large language model can also be used to automatically generate documentation related to the code it generates, such as automated test scripts, which have long been a time-consuming task for automation engineers.

At Schneider Electric, for example, we’ve been testing specialized versions of large language AI models and applying software for EcoStruxure machine experts to train these models. The test results show great potential – code can be generated quickly and accurately, although human review is still required. We can see that the same model can be applied to other software applications.

2. Natural language interface

Large language models also create natural language interfaces for industrial automation systems, so that operators can interact with these systems using only human languages, without the need for specialized programming languages. One of the key applications of natural language interfaces is the use of natural language commands to access existing documents, such as technical product information. In many industries, operators and maintenance personnel must manually search for information in product documentation. Some of them are even paper documents! However, as documents are digitized and transferred to secure specific large language models, these tools can be used to quickly ask questions and find answers. For example, “What is the meaning of error code 8975?” How to fix it?” This kind of question can be easily answered. The key is to use a specific large language model that is secure. These models should only extract answers from audited and officially available manuals, technical notes, and source code.

A common challenge faced by industrial companies is that industry knowledge may only be held by a specific group of people. This has changed, though, with the digitization of assets, the digitization of working practices, and the changing demographics of the workforce. But in many factories, there are still situations where machines or assets fail and the only person who knows how to fix them is not on duty. Imagine how much time and productivity would be saved if on-duty maintenance personnel could troubleshoot, access documents, and (with security measures in place) fix faults with the help of artificial intelligence through voice commands?

Virtual assistants for technical support are another application scenario where natural language interfaces with large language models can bring value to industry. Industrial companies that provide highly specialized and customized support to customers over the phone, such as machine manufacturers working with industrial end users, can take this invisible knowledge and feed it into specific large language models that virtual assistants can leverage to enhance the customer experience and effectively reduce problem-solving time. The same applies to the creation of user manuals and documentation for machines, automated products and systems. Creating these critical documents using a large language model saves time for application engineers, allowing them to apply their knowledge and skills to developing better machines, products, and systems that improve quality and reduce time to market.