Soon it will be possible to connect everything. Components and machines are being equipped with additional sensors and higher performing controls. This means more data is available in manufacturing.

The mountain of data is growing. But what use is it? Manufacturing data should be a means to an end.

The manufacturing information we collect must create added value. How can this succeed? The right data must be collected, connected together intelligently, analysed and the right conclusions drawn, as quickly and as comprehensively as possible.

The information technology (IT) industry shows the way: suggestions in search engines, webpages that offer related articles, language detection in your smartphone are all things internet users see every day.

This isn’t magic, it’s mathematics. Algorithms learn relationships, group users and then generate suitable suggestions.

Machine learning makes use of self-learning software to process enormous amounts of data in the shortest time possible. A widely used approach is to use artificial neuronal structures that imitate the human brain. In this way complex relationships can be recognised, learned and reproduced, where rules-based “if/then relationships” only scratch the surface and simple programmes fail.
 

NEW KNOWLEDGE FOR COMPLEX MANUFACTURING

For the manufacturing industry, this method contains enormous optimisation potential. For example, it can be used to more easily plan complex manufacturing processes, reduce the energy consumption of machines or increase machine availability with preventative maintenance. Currently, automation technology companies offer solutions that allow operating statuses to be visualised online and make different kinds of data available. But this is just a first step.

With machine learning, software can detect wear patterns and its effects on manufacturing systems.

Bosch Rexroth already uses this method for preventative maintenance of systems with very high standstill costs, such as in steel plants, paper factories or offshore installations. Numerous sensors in and on all critical components and modules continuously record operating data. This data is sent to a cloud where status information is calculated, drawing on data from all connected systems. With every data set, the software is better able to recognise patterns and the prediction accuracy for wear and potential failure increases continuously.

KNOWLEDGE OF MACHINES A REQUIREMENT

Creative programmers with machine learning and domain knowledge, so-called data scientists, need to transform mechanical engineering and process know-how into a new generation of software. Without the knowledge of which kinds of components the data comes from, or when maintenance was performed on a machine, for example, the software cannot produce any useable results. The principle is self-evident: the more comprehensive the data set, the greater the potential use for all involved.

With cloud-based machine learning, data protection plays a decisive role, since Industry 4.0 can only be put into practice with the trust of customers. This is an important requirement in order to profit from new findings obtained from the data. 

At Bosch Rexroth we take great care with the data of our customers and we explicitly seek approvals for which data we can use for which purpose. Processing of the data is done on Bosch servers fully equipped with the latest, state-of-art data protection capabilities.

(The author is a member of the executive board with responsibility for the business unit Industrial Applications and Coordination Sales at Bosch Rexroth AG).