Machine learning in automation

In the 21st century, we no longer try to turn lead into gold, but data into money. Machine Learning therefore becomes our modern Philosopher's Stone. While the search for the Philosopher's Stone occupied researchers for 2500 years-leading to porcelain and not to gold-we are optimistic to achieve better results in a shorter period of time. Machine Learning has been an established field of research for decades, yielding a large set of tested algorithms and many successful applications. Nevertheless, the application of these algorithms to Cyber-Physical Systems (CPS) is a rather new field which still poses several challenges. First of all, CPS require solutions which fulfil nonfunctional requirements such as safety, security, maintainability and real-time capabilities. They also show a complex timing behavior which is often due to complex causalities and due to state-based system behavior. Furthermore, the number of useful data points is rather small compared to other domains, mainly because the observed data is highly redundant and covers only a few typical normal situations and not much more informative marginal situations. Several solutions to these challenges are currently in the focus of research, e. g., the integration of a priori knowledge into Machine Learning algorithms (i. e., Greybox-Approaches), the automatic configuration and choice of suitable algorithms (i. e., AutoML), or the development of specific algorithms. However, questions such as data quality, big data platforms and time synchronization also play an important role. This special issue will highlight some of these research trends. The question of interpretability is in the focus of the first paper, ""Interpretable machine learning with reject option"". The next paper, ""Combining expert knowledge and unsupervised learning techniques for anomaly detection in aircraft flight data"", deals with the integration of a priori knowledge on the learning methods. The papers ""Computation of energy efficient driving speeds in conveyor systems"" and ""Online parameter estimation for cyber-physical production systems"" outline methods to learn models for the purpose of prediction and optimization. Finally, the papers ""Simulation and optimization of industrial production lines"" and ""Assessment of variance & distribution in data for effective use of statistical methods for product quality prediction"" apply learned models to prediction tasks.