From Legacy Systems to Smart Factories: How Siemens’ is Guiding Manufacturers Toward Industry’s Future

Second, a lot of the data and information in these companies is spread over homegrown systems, which require quite a bit of attention. For example, we have a customer that implemented a Siemens quality control system nearly 30 years ago. That system is still working, but it now has 60 interfaces to local systems the customer developed. Unfortunately, the people who developed these interfaces are no longer with the company, which makes switching to a modern quality system a big risk because it’s difficult to determine which of these interfaces is still relevant and may connect to an active and important system. 

The third domain of challenges relates to how you create a digital twin of a brownfield factory. To do this you need to model the factory. The challenge here is to identify a way to capture large amounts of unstructured information to identify all the equipment and connections that exist. We are working on how to use AI agents to run through this scanned factory data, capture the silhouettes of the machines on the production line, and then look for information on the network to discover the correct equipment versions and model them for a realistic digital twin.

AW: How does Siemens define the comprehensive digital twin concept and what differentiates it from a more standard digital twin technology approach?

RG: For Siemens, the comprehensive digital twin includes building a digital twin for the entire lifecycle of the machine. When you're creating a concept of the machine, you create a digital twin of the machine at the front end. And that concept could be developed using 2D design schematics to capture, say, the type of pump, actuator and electrical power used. 2D renderings can be used to create a digital twin because they allow you to evaluate the pumps and the compressors and the forces being applied. From there you can create the electrical schematics that will be part of the digital twin as well as the automation design. Then once you have everything designed, you can create the manufacturing blueprint for how it will be produced inside your plant. The comprehensive digital twin provides a complete digital twin of the humans and robots working in that factory. Even the conveyor systems that will connect to that machine are represented.

We recently worked with a packaging machine company on a lettuce packaging application where the machine had a shaft component that moves up and down. Sometimes that shaft component would get jammed, so they wanted to design a safe way for an operator to open it and clear the jam. This old access point was very heavy, so they wanted to redesign it so that anyone could easily unlock and open and close it. They did this redesign and testing with a digital twin, which allowed them to address the design for use by short or tall workers for ease of deployment anywhere in the world.

That’s what we mean by the comprehensive digital twin. You can test every part in the machine — every PLC, every motor — inside the digital twin, giving you the ability to evaluate and test the design of the entire machine and evaluate every function of it in the factory. 

AW: Since industry applications of any technology differ widely, explain Siemens focus on the out-of-the-box capabilities of Teamcenter X and how that can be used to address the specific needs of various industries like aerospace, automotive and consumer goods?

DT: When you think manufacturing simulation software, many manufacturers are buying the same solutions, but how they apply it is slightly different. For example, an aircraft manufacturer may apply it in a certain way to build 500 aircraft a year, versus an automotive company that's building 500,000 cars a year. So, of course, there’s tailoring of those solutions going on in these different applications. But the more you can have that out-of-the-box functionality, it's easier to tailor it through configuration rather than coming up with a unique solution. The benefit of this is that, when we add functionality to solve the needs of one industry, other industries can benefit from that right away. 

Whether you're talking about discrete manufacturing like aircraft and cars or process manufacturing like food and beverage or pharmaceuticals, we're able to leverage a broader product portfolio to help both of those industries with a lot of manufacturing commonality with these out-of-the-box solutions. We recognize that each industry has unique needs and business challenges and regulatory requirements, but at the end of the day, how they produce their products are similar. And the more we can leverage those similarities, we can provide more value to those customers by thinking about this with a cross-industry point of view to provide that out of the box.

AW: The focus on digital threads is gaining interest again with wider applications of generative AI co-pilots. How is Siemens leveraging digital thread technology centered around BOMs to streamline the lifecycle of industrial machinery products?

RG: Digital threads are the glue that enables a company to work faster and minimize all risks. The packaging company I referenced earlier with the lettuce wrapping machine, they also produce a machine for cereal packaging. They went through years of work and efficiency changes to reduce their delivery schedule for these machines to 28 weeks. Their CEO now wants to get it down to 12 weeks. The only way they can do that is by improving the efficiencies inside that domain while also improving operations across the company lifecycle.

The biggest challenge here has been a disconnect between engineering and manufacturing. For example, to improve the machine’s design for faster production, the engineers focused on a plate that controls how cereal pieces flow into the box. That plate has a special size dimensioning that varies depending on the size of the cereal pieces — its curvature and angle changes depending on the cereal type being packaged. The problem was that the design process used to create that plate and the manufacturing process used to install it were totally disconnected. The company didn’t have a good way to send information from the design engineer to the manufacturing engineer. They were just sending files to each other and anytime the manufacturing engineer saw a problem, they would send a message back to the design engineer. 

This led the design engineer and the manufacturing engineer to go back and forth trying to understand the problem. That process often took a week. Then the design engineer would fix it, but that took another week. So, the company would lose two weeks for any change process. 

These situations happen everywhere — even in the order bidding process. For example, a manufacturer will find another carboard supplier for their cereal boxes that’s better quality and weighs less. That weight reduction has an impact on the way folding is done in the box-producing machine. It effects the number of folds that can be put on the box and the speed at which the boxes can be folded. It all ripples through the whole engineering and manufacturing processes. Here again, communication about these changes gets routed between the various departments, and that information flow can break down inside a company.

With a digital thread connecting everyone and every process and product, when a design engineer comes in on Monday morning, he doesn’t have to look through emails to see what he may need to do. Instead, any designs he needs to work on are flagged in the system. And until that task is completed, that flag will be there. That’s why more manufacturers are recognizing the value of the digital thread and why I say digital threads are the glue that can keep all these communications and changes together in a consistent, seamless manner so that information doesn’t slip through the cracks. 

AW: Data management has been a major digital transformation issue, but what else should manufacturers be focusing on, particularly for AI applications?

ZF: I think the topic of data is overemphasized. Of course, data is important, but I’ve been dealing with AI topics for 30 years now and, in the past, we may have needed millions of pieces of information. Today, we can teach an AI large language model (LLM) to work even better with just 200 pieces of information. So we need to be smart about how we define what data we need and use RAG (retrieval-augmented generation) to augment the data and bring it into a large AI model in a way that it will not drop all your intellectual property into the cloud. 

To figure out what data you need for AI analyses, I recommend the Pareto rule. With AI data analyses, this means that 20% of your data enables 80% of your success. At Siemens, we are developing an LLM that addresses data privacy aspects and can be used to develop 3D objects for digital twins. Here, we are focusing on linear models for machine parts — these are components that most LLMs cannot work with yet.