Artificial Intelligence

Machine learning is the key technology of almost every instance of modern Artificial Intelligence. Enormous datasets are produced in digitized industrial processes and in the Internet of Things, which can well be exploited by learning in deep artificial neural networks. Standard machine learning algorithms require these datasets to be centralized before learning a model. Several good reasons - ranging from data privacy over latency to economic efficiency - favor learning at the edge so that reasoning is fast and no local data is transferred. The article shows how decentralized learning works and how to evaluate it. Moreover, we point to special resource-efficient learning algorithms and discuss small remaining risks of data reconstruction.

Load bearing structures are typically designed with respect to relevant load cases, assuming static shapes and given material properties that are selected during design and material selection. An example of such a material is a particular class of polymers that are capable of changing their elasticity based on the influence of optical, thermal or electrical fields. One problem to be solved in terms of active intelligent cellular structures is the correlated and self-organizing control of cell response and control and the underlying information organization that must provide robustness and real-time capabilities. We propose a hybrid approach combining mobile and reactive self-organizing multi-agent systems (MAS) and machine learning. The MAS is the essential robust information communication technology (ICT). The agents are executed in material-integrated networks consisting of microcomputers. The simulation and implementation of such complex systems is a major challenge.

Wire ropes are used in different applications and human life often depend on their integrity. Therefore, technical personnel checks the tightropes on a regular basis but there are some difficulties in detecting damaged areas. Consequently, wire ropes are exchanged rather too early than too late causing avoidable extra costs. In this paper, the project MOBISTAR is presented that combines a magneto-inductive and an optical sensor to detect damages and a software based on Convolutional Neural Networks to evaluate those defects.

Industrial devices have virtual and physical components that interact with each other in a plethora of ways. We report on a system that enables operators to pose queries about physical, regulatory and functional relationships between components and visualizes responses as a holographic overlay, thereby enabling in-situ querying and rendering of information for “on-the-spot” decisions. Importantly, our approach is not only applicable to digitally integrated components but applies equally well to “simple” objects such as surfaces and workpieces, and the materials they are made of.

A complex process in sheet metal processing is multi stage deep drawing. Once set up, it can be considered as a black box. Usually, after the sheet metal has been processes the quality is assessed. In the research project Smart Press a system is developed, incorporating inline pictures of the processed sheet metal. Pictures of failures are related to the actual state of the machine. Neural Networks are used to model the highly complex relations between parameters and product attributes. Based on the assessed real time data, the process gets adjusted to suit the needs of each individual sheet.

Technological progress offers opportunities for introducing new business models but also opens new paths for attackers. Security-by-design integrates factory security from the start and introduces a continuous security process built upon a variety of components. This article illustrates some of these components by highlighting secure processes with specific case studies: anomaly detection within value networks and secure production planning with simulation.

The digital transformation led to disruptive changes in business models of leading companies. Big Data serves as one of the key enables in this area. The transfer of this concept in the production domain towards an Industrial Big Data is key challenge for producing companies. Although exemplary key projects exist, no available characterization of structural elements in Industrial Big Data Processes exists. Therefore, this article aims at presenting initial structural elements of Industrial Big Data projects based on exemplary use cases.