Supply Chain and Engineering: Generative AI LLMs in Rolling Bearings.

The world of engineering is rapidly changing. Traditional processes are being overturned by groundbreaking advancements. Generative AI and LLMs (Large Language Models) are disruptively redefining engineering and design processes. These technologies are making specialized knowledge, once exclusive to traditional manufacturers or a limited number of experts, accessible to everyone, transforming roles and redistributing competencies.

There is a strong demand from multiple sources for a more connected, agile, and smart supply chain, empowering the industry to innovate faster and more reliably.

McKinsey, for instance, in their report "Deep Learning in Product Design" explains how new technologies are transforming the way companies approach product design and engineering. They highlight that the overall process has become substantially faster, with a potential further 50% reduction in time to market, a 25% improvement in quality, and a 10% revenue uplift when new technologies are used to enhance customer experience.

For about five years now, within their "Megatrend" analysis about digitalization, Siemens has been talking about "Zero Engineering" envisioning a world of automation where engineering practically takes care of itself. They are actively building a world where machines and systems autonomously learn, understand, interpret, and make decisions within defined limits.

Rolling bearings are a rather mature technology, engineered for over 100 years and consisting of a relatively limited number of simple components. Yet, they are among the most heavily loaded mechanical elements, often defining system performance, reliability, and the lifespan of machines. As such, they appear to be the perfect component for the application of conversational Generative AI LLM agents. Given the abundance of publicly and proprietarily design and performance data, collected since decades, and readily available for training, rolling bearings are well-suited to benefit from the interaction between human and artificial intelligence.

·?????? Design rules and principles are relatively easy to define and have been well-known for decades.

·?????? Clear approaches and guidelines have been reliably used for decades for the integration of rolling bearings into assemblies, subassemblies, OEM solutions, and final systems.

·?????? Defined and collectively accepted guidelines have been established for decades for the selection of optimal bearing solutions for given applications.

·?????? Dedicated industry standards are established and universally accepted across the globe (or almost!).

·?????? An abundance of field-collected data describing their static, dynamic, thermal, and vibratory behavior has been collected for over 20 years in structured and unstructured databases by both bearing manufacturers as well as OEMs in all industries

Despite all the above, these changes are often resisted, overlooked, or sometimes unnoticed, even though they have a significant impact when applied within the supply chain, enormously simplifying the way engineering interacts with it.

In the following section of this article, we will present one of many practical use cases.

Use Case – A Buyer

Imagine being a buyer in the purchasing team of an OEM. You are not an engineer or a technical expert. While reviewing the project documentation, you come across a bearing specification for standard ACBB 7204. You are unfamiliar with bearings and unsure if this is still the best solution for your project, as the documentation is a few years old and newer solutions may have emerged

You don't know exactly what to purchase, especially since distribution and purchasing channels have rapidly changed over the last few years and continue to evolve. Is a top-brand necessary for your application, or would a lower-priced supplier suffice? How do you make these determinations and come to a decision?

Initially, you start by calling your engineering department to ask a few questions. However, as we all know, engineers are always very busy and might not have the time to discuss your bearing needs right away. Even if someone in your engineering department becomes available, they may not be knowledgeable about bearings. They might need to rely on a subsequent call to a bearing manufacturer's application engineer to explore possibilities and understand solutions.

Once again, however, application engineers among bearing manufacturers are also very busy, often not immediately available, and with their own project priorities not always matching your expectations.

Let’s be clear: ultimately, they must be involved in the final decision-making. No doubt about it! However, there is still a lot to do and a lot to understand before reaching this final decision point.

The buyer needs to know the specifics of the bearing: its basic working principles, typical applications, specifications such as load-carrying capacity, speeds, dimensions, and if available, executions with seals or shields. The buyer wants to pre-screen if other solutions should be considered or compared to this one. A basic life calculation might be advised to make a preliminary determination if the bearing is still a suitable solution, whether it is called to work with large or low safety factor, ultimately leading to the decision between costly top-brand products or lower-priced solutions.

Ultimately, this could lead to a project review with higher involvement of engineering, reconsideration of application inputs, a potential design review or upgrade, and possibly a newer innovative solution. However, the real question is: how do we reach this point quickly enough, with already prescreened options and involving the relevant expertise only and strictly when required?

What are the available solutions today to carry out these screenings in a simple and fast way?

General Catalogues and simple calculations

The buyer can decide to look up manufacturer general catalogs or specific online platforms to search for a 7209.

This would help to understand the bearing type, basic specifications, typical executions, and other versions such as sealed or shielded bearings, as well as information about typical fields of application and lubrication choices. Some searching and reading would be required to gain information about typical angular contact bearing arrangements, ISO tolerances, and fitting conditions on shafts and housings.

All of this is not to come to a final decision but rather for gathering general information and perform an initial screening. Considering that this task is undertaken by a non-engineer, non-technical person, we are already looking at a few hours of time investment in this endeavor, to say the least!

When it comes to initial screening calculations, (assuming in this example that a basic life calculation must be performed), the buyer must gather information about the basic rating life model, understand how to perform the calculation, use the bearing specs from catalog tables, and make a simple calculation. Although relatively straightforward for engineers, it is still very difficult for a non-technical person to judge the outcome: is the life too long or too short? In fact, one can say without the risk of being contradicted that all of this rarely happens!

In the real world, the buyer would wait until the internal engineering department becomes available or until an application engineer from the bearing supplier is willing to provide support and spend some time. This enormously slows down the design, review, and often the innovation process, which could otherwise be achieved and accelerated if the right support tools were used!

Simple calculation tools

An alternative, more viable option would be the use of simple calculation tool available on countless platforms, including bearing manufacturer software. These are all great tools, extremely powerful and precise, offering suitable ways to make calculations. However, they are all designed for technical/engineering people. One of the best and more simple solution is shown in the picture below.

Regardless of its simplicity, for the buyer to be able to perform a calculation, a significant investment in time is required. The buyer must know how to use the tool, how to set it up, how to define boundary conditions, how to input application data, how to run the calculation, and how to document the result. Additionally, some knowledge and experience would still be required to make a final judgment on the outcome. This approach, while more likely to be adopted, still requires several hours of time and/or significant involvement of internal or external engineering expertise.

Generative AI: A Virtual Bearing Engineer

Generative AI is being deployed as conversational agents in several sectors. At the Hannover Messe 2024 in May, several applications were showcased where the technology was used to enhance digital customer experiences in product engineering and smart supply chains. The majority of deployments are focusing on augmenting knowledge by leveraging extensive libraries and datasets. For instance, Kone uses this technology to empower field technicians and enhance customer service, while BMW is augmenting regional supply chain management by enabling non-technical users to generate sophisticated knowledge and analytics.

According to Forbes, by 2026, more than 80% of enterprises will have used generative AI to transform their business models and customer offerings.

As an example, in this chapter, we'll see how a generative AI LLM can assist in our use case, helping the buyer.

A well-trained LLM can assist indeed our buyer in a matter of minutes. Generative AI has the potential to provide all the knowledge required to perform these screening tasks. This expertise, once exclusive to a few experts and specialized teams, is now democratized by the application of generative AI and becomes accessible to everyone.

In order to grasp fully the potential of the technology, imagine a fine-tuned LLM (Large Language Model) working on a dedicated platform available to customers. The platform (working as a Virtual Bearing Engineer) guarantees the utmost level of confidentiality, ensuring that proprietary data is not shared in any form. This allows customers to share their information freely and confidently.

The Virtual Bearing Engineer operates via API calls and functions as a conversational agent trained specifically on the subjects of rolling bearings and their basic calculations. It includes a retrieval module, an ingestion pipeline with a dedicated RAG (retrieval-augmented generation), and specific preprocessing and data readers, embedding generation, and data sink writer. This system is designed to handle not only text but also tables, figures, sketches, and plots.

The system is further enhanced with mathematical function callings. The calculations are not simply delivered by the statistical approach typical of LLMs, but through defined function API calls.

In a few seconds, the buyer understands what a 7204 is. In the picture below, you can see a simple example of querying a virtual engineer. (The shown content takes literally 4 seconds to generate).

By referring to the project documentation data, and providing the virtual engineer with standard inputs, the buyer can perform a basic life calculation within seconds and without needing to know how to use a simulation tool. By inserting application data in a conversational way, a life calculation can be reliably and consistently performed within seconds.

Additionally, within a couple of minutes, the buyer can gain a decent understanding of the outcome of this screening. How long is the life? Are there options to extend it further? Are there other bearing solutions to consider? All these questions can be preliminarily answered by the virtual engineer.

Conclusions

Make no mistake, though. We do not believe that this technology is ready to always make fully independent decisions without human input. Not at all. We are still far from this point. However, the screening of options, simulations for pre-project assessments, quick checks of different scenarios during the divergent phase of a project, and more can be remarkably accelerated.

Moreover, a larger group of employees can now perform tasks that were once exclusive to very technical profiles. This will surely accelerate the process of development and engineering, ultimately making the innovation process faster and better.

Not only that. Imagine a virtual engineer providing purchasing information such as where to buy, delivery times, prices, availabilities, alternative options, lower-priced solutions, and second brands.

Taking it a step further: envision a buyer using a specific bearing manufacturer's virtual engineer, guiding them through its own catalog, products, and calculations. This would make the interaction between bearing engineering and the supply chain seamless, fast, cheap, and effective.

We have no doubt: the world is changing fast. Innovators, forward-thinking organizations, and companies aiming to lead their industry have an enormous opportunity to embrace these changes and be the first to adapt. Corporations seeking to protect their leadership position must also quickly embrace these new paradigms in engineering and value creation by democratizing knowledge within their customer base.

?

?

?

?