Data Science Series, Part 1

“Data is the new oil” is one of the most frequently heard sentences in the age of digitization: Data is increasingly becoming a company’s most important economic asset. This also applies to mechanical engineering and industrial production. As examples from Körber Digital show, applying Data Science is less complicated than many people think. It all depends on the right expert support.

Increasing performance? The data response is already available!

Today, data-based companies are among the most valuable in the world: they design their processes particularly efficiently, understand their customers much better and develop their products on the basis of data. It’s a little surprising that industrial companies and mechanical engineers are progressing cautiously in the increasingly important field of Data Science. After all, machines and systems generate huge amounts of data that can be used to increase the productivity of a process or the manufacturing of a product. For example, only 8% of mechanical engineering companies in Germany collect, analyze and use large amounts of data, according to a Commerzbank study from 2018. Nevertheless, the companies surveyed expressed clear goals through the systematic use of data. They range from better decision-making and better utilization of resources to the development of new business models, products and services.

Systematic Data Science

Targeted data analysis and data modeling can only start once the purpose of the Data Science project is clearly defined by all participants. Later, the results are presented to the customer in detail, the required implementation is carried out and finally the outcome is measured and documented. But how can you envisage this process in practice and how great is the benefit? There are impressive examples of this. For example, Körber Digital has optimized material handling for a customer using an industrial gripper – without investing in new production technology. Instead, the existing industrial gripper was given a smart “update.” Now it picks up various components more sensitively than before and guides them further. The Körber Digital specialists even familiarized themselves with the electromechanical processes of the gripper and comprehensively visualized the data of a gripping process. Now the gripper can not only automatically detect whether a gripping process has been successful, but also assess whether a component has been damaged or not.
“Our Data Science approach is interesting for all those who are already collecting data and now want to know what problems they can solve with this data,” says Warnke.

Applied machine learning: cow health in the data focus

A Data Science project at a mechanical engineering company producing milking machines is similarly impressive: Together with the customer, Körber Digital has developed a system that determines the health of cows using their milked milk. Sensor measurements on milk temperature, milk quantity or milk conductivity are analyzed using machine learning algorithms because there are very complex relationships between this data and udder health. An automated analysis of cow health is now carried out – so up-close visual inspections by the farmer are no longer necessary.

Outlier detection

How are potential “false alarms” excluded here and in other areas of application for Data Science? This is where the fundamental topic “outlier detection” comes to the fore. It underscores what is important for every Data Scientist: “understanding” the data. After all, the mere collection of data does not generate any added value itself. Rather, clear recommendations for action must be derived from this. In this context, the mentioned outliers are important. Put simply, these are data points that do not meet expectations. In order to “detect” them, Körber Digital experts measure the distances between two of these data points and classify outliers. As a result, a complex data cleansing process is necessary before one can say with certainty whether there is really a technical problem – an “anomaly” – or just a random measurement error. “With the help of secured anomalies, we can identify defective components, for example, or discover negative effects caused by retrofitting a machine,” says Dr. David Breyel, Data Scientist at Körber Digital. “The possibilities are very extensive.”

Now is the time to use Data Science

In the end, the central question remains as to which mechanical engineers or manufacturing companies are suitable in general for a Data Science project. “The answer is manifold,” says Warnke. “For example, our approach is interesting for all those who are already collecting data and want to find out which problems they can solve with this data.” The same applies to companies that use intelligent machines or sensors, whose production has a high degree of automation, or those that can already (partially) digitally map their production process. “We are also frequently called in when a company has initial ideas for a new innovation and wants to evaluate its suitability using existing data. This is where Data Science provides valuable insights,” says Warnke. “There is no doubt that comprehensive Data Science will become an elementary and indispensable part of mechanical engineering in the 21st century.”