AI and the manufacturing cycle train

⇨ SAP emphasizes end-to-end integration to streamline manufacturing and improve efficiency.

⇨ Advanced AI and IoT enhance maintenance strategies and predictive capabilities for manufacturing.

⇨ SAP leverages AI and data for faster, smarter manufacturing and product development.

Providing critical services to consumers and customers around the world, organizations in the manufacturing industry are always looking for ways that help achieve smoother, faster and more efficient means of production. This is especially crucial in light of a variety of challenges faced by companies today, such as supply chain disruptions, labor shortages and geopolitical uncertainties.

At the Hannover Messe 2024, SAP executives presented insights and updates on where the company is heading with its solutions for digital supply chain in manufacturing and how this all ties in with its AI strategy. Georg Kube, head of industry data ecosystems, and Dominik Metzger, head of SAP digital supply chain, share SAP’s most recent updates on the transformation of the manufacturing cycle and how they can improve users’ experience.

Make the best of your data collaboratively 

“Manufacturing is one of the key value drivers globally. It is a point of contention between nations. It is a thing that creates wealth, value and prosperity for everybody,” says Kube in the opening statement.

This year’s three major priorities for SAP that Kube highlights include the emphasis on end-to-end processes, the use of data and the recognition that no customer succeeds alone.

According to SAP, the emphasis on end-to-end processes is important because, if customers were to consider what the traditional experience of implementing manufacturing systems is like, they would know that integration is not common and all functions that are fundamentally tied within the production process are dispersed and handled separately. Today, there is a new approach to end-to-end processes, in which manufacturing functions, including development, supply chain and customer interface configurations, are all Integrated. 

The second priority – the use of data – is guided by both products and customers becoming more data-rich. The processes that create products also require data to be optimized from a quality perspective, speed and other factors. Finally, SAP highlights the need to recognize that “no customer succeeds alone.” Specifically, that new solutions will support customers’ efforts in collaborating in ecosystems and embedding Networks.

The three major priorities described by Kube set the direction for how exactly SAP is going to streamline manufacturing processes, as explored by Metzger next.

Hop on the manufacturing cycle train: first stop, recipe formulation process

 “It all starts in the engineering or recipe formulation process”, says Metzger. Having chosen batch productions as an example of the industry that the updates will affect, he describes how SAP is going to “significantly accelerate” the idea-tomarket process for companies.

Batch productions (which include such businesses as pharmaceuticals, beverages, chemicals and others) manage thousands of customers’ product data in their ERP systems, including very granular descriptions of every single ingredient. So, “In order to accelerate product incubation”, Metzger explains, “we need a lot of data beyond the engineering department or the recipe formulation specialists.” 

So, for SAP, the next most important aspect is collaboration in the ecosystem. “Sometimes as a manufacturer, I don’t have all the ingredients details, such as information on CO2 emissions of a certain product that I want to incubate or a certain ingredient that goes into my product.

“What we are enabling with our Product Lifecycle Management (PLM) is a deep integration process of that data into the SAP Business Network so that users can digitally request any data from suppliers that they require to create a sustainable and correct product for the target market. Even here, collaboration with suppliers is key.”

Another way to accelerate the idea-to- market process is AI. According to Metzger, this usually involves a lot of “creative cycles.” Product managers come up with product ideas, such as what can be launched to market and what campaigns can be organized.

However, Metzger explains that all ideas and customer requirements need to be rationalized. Large language models (LLMs) for PLM, especially for this early ideation and product requirements management, is the answer to that. 

Continuing the journey: How to streamline the shop floor production 

Once the product is designed, it needs to be manufactured on the shop floor. The primary obstacle in the end-to-end process of manufacturing execution that SAP addresses is the lack of agility during it.

In the traditional process, companies cannot change their production cycle without intervening in the entire logistics, delaying the process by, for instance, moving raw materials back to the warehouse or into staging areas. The same goes for if something goes wrong during the process, such as quality issues or defects.

To prevent delays and improve logistical considerations, Metzger says, shop floor executions need to be deeply integrated across the silos.

SAP launched the AI-powered Visual Inspection solution last year, but today the company “has significantly extended the use of visual inspection capabilities to be much more scalable and easier for customers […] so that any SAP data can be used to train a machine learning model to identify right from wrong.”

“What we have developed with the help of the LLMs is meant to significantly accelerate issue resolution,” Metzger adds.

Reaching the destination and improving maintenance

At the final “mental station”, as Metzger calls it, when the product is made, the next step for manufacturers is asset operations and service. For SAP, he says, it is time to define how to achieve a “digital thread” that would receive all the data that was created when designing a product in a full system record in SAP. This is especially the case when manufacturers need to collaborate with the operator of their asset and share such data as maintenance and service instructions or spare parts information.

“What SAP really focuses on now is [the question of ] “How do I come up with the right maintenance strategy in the first place?”, says Metzger.

“A state-of-the-art method nowadays is what we call ‘failure mode and effects analysis.’ I want to know what could go wrong, what failures could occur within this asset, what is the effect of a failure and from that derive the right maintenance strategy.”

Additionally, with SAP’s recent announcement of an extended partnership with Cumulocity IoT, customers will be able to define what the right asset maintenance strategy is by tapping into IoT data to connect their SAP maintenance software for predictive strategy methods. The predictive maintenance will then allow service operators to find out if there’s a spike in indicators and anticipate an asset failure by using anomaly detections.

“Especially in the context of service worker utilization – we’ve implemented machine-learning capabilities to identify more accurate durations based on past service orders. Thus, the machine learning algorithm can get smarter and determine the actual service duration for the workers in the field.”

So, how is this AI different from the “household AI”?

When companies decide to explore the capabilities of AI, a common question arises – what is the actual difference between “business AI” and “household AI”?

The biggest difference, Metzger explains, is in the use of data. While the generative agnostic LLM is pre-trained to use public domain data, SAP is embedding vetted data into the model’s database.

“For example, [you have] a vetted PDF document which may have 200 pages of how exactly [to] use this system, let’s call it our User Handbook. So, we are embedding this into a database and that’s what the LLM executes to answer user’s questions, such as ‘How do I configure the system?’, ‘How do I find or troubleshoot a problem?’ We are not asking a generic model but embedding recipe data that you already have in your SAP system to ensure that you have the right business data backing up the LLM context.”

Another important point of differentiation comes at what Metzger calls “cost value.” Bringing Google’s latest version of Gemini as an example, he explains that, although these are very advanced LLMs, they may not solve companies’ recipe problems. So, SAP combines LLMs to have the ability to solve a single problem through further training and fine-tuning, thus optimizing the models’ engineering from the cost side.

“No models in the past were able to interpret small elements [like this] and compare them against the SAP system but now LLMs will be able to differentiate them. […] Compared to traditional machine learning models where the training element is costly and takes much longer, this is something we will do based on fortified customer data that we are allowed to use at SAP. We will also give this tool to our customers to train on their own data depending on data privacy and other security considerations that they may of course have.”

Although discussions and amazement around AI are echoing around the world, the deep integration and implementation of it in companies’ internal processes is still unfinished business. Understanding how AI can benefit and secure production processes, supply chains and service management in manufacturing can set organizations to a new level of digital transformation. While the updates offered by SAP are just the beginning, future innovation in the AI sphere will certainly keep us on the edge of our seats for a long time.