Marginal Column

Input counts


March 2017


What fills some people with concern, inspires others. Will machines soon be more intelligent than people? The potential of intelligent machines was confirmed when an algorithm beat one of the world’s best players of the Asian board game Go. But how can machines learn? And what is meant by the term “machine learning”?

“You were interested in this product. Perhaps you’ll like this one too?” Be it web shops, advertising, or search engines: computers seem to know what we want. It’s nothing to do with empathy, just mathematics. They track our user behavior, compare it with that of other users, identify matching patterns, and use them to derive likely interests.

The requirements for machines to make logical suggestions, give sensible responses to voice input, or trigger an alarm in case of anomalies are always the same – huge data volumes, enormous computing power, and training. Machines acquire empirical knowledge based on the methodical and systematic collection of data. Algorithms such as artificial neural networks help to structure the data. The more comprehensive the data records available, the more accurately the machine can recognize patterns in complex data and create relationships.

The data available to the machine and the algorithms it uses to achieve the required target functions are down to people. For example, for industrial applications, data scientists – creative programmers with machine learning expertise – have to convert engineering and process know-how into intelligent software. With ODiN Predictive Maintenance, specialists from Bosch Rexroth have developed a monitoring system for predictive maintenance. The system uses machine learning methods to generate knowledge about the health status of a system from the sensor data acquired and to make reliable predictions. Customers then receive the corresponding maintenance recommendations for their systems.

An example involving wear diagnosis on industrial systems illustrates the capability of these systems. Whereas, statistically speaking, detecting a fault by chance will only have a probability of 13 percent, and an expert constantly monitoring the system will achieve a success rate of 43 percent, machine learning increases the fault detection rate to over 95 percent.

Summary: Machines are capable of learning. The prerequisite is that the algorithms have access to comprehensive data records. They can then reliably draw intelligent conclusions that benefit users.

Machine Learning

Machine Learning