Machine learning: thyssenkrupp relies on intelligent algorithms

Ritzmann's career path is remarkable: after studying physics and seven years in quantum physics research, he decided to make a career change. Ritzmann taught physics at the Studienkolleg Bochum, preparing young people for technical studies. At the same time, he discovered his passion for machine learning. He finally took the plunge and dedicated himself professionally to this exciting field as a data scientist at thyssenkrupp.

Mastering complexity with machine learning

To understand how machine learning works at thyssenkrupp, let's first take a closer look at the term. Unlike many other applications of artificial intelligence, machine learning is not about imitating human intelligence as a whole. The aim of machine learning is to teach a model to perform a specific task and deliver precise results by recognizing patterns.

For Ritzmann, the fascination of machine learning lies in designing machines that can independently recognize patterns. This enables them to solve complex tasks – almost as if they were capable of thinking. In this way, machine learning opens a completely new field of possible applications. Especially when compared to traditional manual workflows or classic algorithms based on precise logic and fixed rules. Machine learning is particularly helpful for problems so complex that we cannot solve them using our intellect and purely formal logic. "Machine learning dispenses with this logic and instead develops a kind of intuition or 'gut feeling' for the solution based on large amounts of data," explains Ritzmann.

The advantages of machine learning: intuition and efficiency

"This type of intuition or 'gut feeling' enables decisions to be made faster and more efficiently than would be possible with manual processes," says Ritzmann. This is a decisive advantage, especially in today's world, where companies are confronted with an ever-growing flood of data. And machine learning is also attracting a great deal of interest at thyssenkrupp: The members of the thyssenkrupp AI Community meet once a quarter. "More and more colleagues are getting involved in the topic. Interested minds are always welcome to join us," says Ritzmann.

Machine learning in practice at thyssenkrupp: the mlATP project

Machine learning is already being applied at thyssenkrupp. For Ritzmann, mlATP is a particularly exciting project. It was developed and put into operation together with thyssenkrupp Schulte. What makes it special? With mlATP, the expected delivery date can be predicted and displayed directly to customers in the web store. Only when they decide to purchase is the planning process initiated in the ERP (Enterprise Resource Planning) system. "The machine learning model is based on the linking of historical delivery data and current stock levels," explains Ritzmann. Currently, the correct calculation of the delivery time also includes planning of all processing and transportation steps. This is very complex and requires a certain amount of time. This is where mIATP comes to the rescue.

Julian Ritzmann and his team have big plans for the future. They want to further develop the mIATP they invented and thus take further steps towards more efficient working. “We are currently working on another project that will enable more efficient planning of customer deliveries. It could collect all orders for a day and create the most efficient and cost-effective delivery plan possible on this basis. mlATP would give customers a delivery date for the products in advance,” says Ritzmann. In this way, mIATP could help to optimize the supply chain and inform customers about the delivery timeframe.

Challenges in data processing

To utilize machine learning as precisely as possible, a large amount of data is required to train the machine learning models. Acquiring this can be a challenge. "This also applies to understanding the data. It is important to know the origin and original use of the data," says Ritzmann. That is why he considers it essential to be in regular contact with all experts and users. "When reviewing the data set, you are often surprised to discover which errors have crept in over time. Sometimes data has also been manually overwritten afterwards," he explains. The only solution to this challenge, according to Ritzmann, is to meticulously check the data set and remain in constant communication with all involved parties.

Another challenge: the fast-paced nature of the field. This requires a wide range of action and short decision-making paths. "If I make a list of all the projects, it quickly becomes outdated. There are always new ideas." But this is precisely where Ritzmann also sees an advantage: "With machine learning, we suddenly have a very big hammer in our hands, and now we just need to find the right nails to hit."

Ritzmann has one advice for those interested in machine learning: "Have fun with it! Experiment with small projects, gain initial experience and show your enthusiasm for the subject." Especially as jobs continue to evolve through digitalization, continuous learning is essential.