The research project Automated Commissioning by Persisting Expert knowledge (AIPE/ACPE) aims to automate parameter optimisation of manufacturing machinery by combining expert knowledge and artificial intelligence. The project is funded by the German BMWi and developed together with the XITASO GmbH. During the project we plan to design a generic architecture/process that can be applied to various industrial processes. As (labelled) data is often only sparsely available we aim to require as little data as possible which will be realised by using knowledge of domain specific experts. Besides the initial configuration of the machine (commissioning) our system will be able to also enhance reparametrisation efforts. Those typically arise after product changes, geographic relocations or (seasonal) climate changes.

For first evaluations of our concepts we use multiple FDM-based additive manufacturing machines (colloquially referred to as 3D printers). We aim to combine expert knowledge with state of the art data driven machine learning algorithms, therefore we first need to develop a method to collect existing „intuitive“ knowledge from experts in the domain and prepare it in a form from which a ML algorithm can learn. To realise the data driven part of our system we will need to systematically collect data in different situations on which the system can learn (enhanced by the knowledge extractions).

Contact

PredMain – Entwicklung von Soft- und Hardware-Frameworks mittels innovativen Predictive Maintenance Algorithmen und Machine-Learning-Verfahren, für die Belastungserfassungen in Fitnessgeräten

FMLA – Forum Maschinelles Lernen Augsburg 

OSCARD – Organic System für Cyber-Attack Recognition and Defense; Selbstorganisation und Lernen in OSCARD  .

CYPHOC – Absicherung von Cyber-Physical Systems mit Methoden des Organic Computing.

Abstract

Cyber-Physical Systems (CPS) connect two quite different worlds, the world of embedded systems (with real-time requirements, sensors and actuators, dependability, deterministic behavior, etc.) with the world of digital networks (with globally available services, data clouds, multi-modal man-machine interfaces, etc.). CPS are exposed to different security threats, many are not known at the design time of a CPS. In general, the physical surrounding of the CPS may be endangered, but also the components of the CPS or the communication between spatially distributed components, for instance. In the CYPHOC project, we address these security problems by means of Organic Computing (OC) techniques. OC focuses on adaptive technical systems, typically empowered with learning abilities, to solve complex problems. Properties such as self-learning, self-adaptation, self-coordination, self-organization, or self-healing play an important role. In CYPHOC, “security-by-design” is complemented by “security-at-runtime”, that is, the components of CPS are enabled to detect new kinds of security threats collectively and to react accordingly. In particular, solving this involves three different research topics: collaborative detection and learning of conspicuous situations (group of Prof. Sick), generalized mechanisms to react appropriately on unanticipated situations (group of Prof. Hähner), and guaranteed protection against compromised components (group of Prof. Wacker). Specifically, we substantially improve techniques that enable CPS to detect conspicuous and suspicious situations in their environment (in particular temporal anomalies) that are not known at design time of the system. Based on the recognition of unanticipated events, we require standardized mechanisms to react appropriately in a self-organizing way. The set of possible strategies to react on these anomalies is too large to be efficiently searched. In many applications, however, dependencies between components exist. By automatically detecting and modeling these dependencies, we can exclude such strategies that do not respect them. Therefore, such dependencies are exploited to realize a faster collaborative learning in different classes of applications. Since most CPS are distributed systems, some components of the overall CPS might be compromised by an attacker. To guarantee protection against such compromised components, we develop mechanisms allowing for any piece of information to be k-resilient. Therefore, an attacker is required to manipulate at least k different components to achieve his goal. Additionally, we investigate the realization of CPS-wide self-tests to detect these compromised components. We design all these developed OC techniques in such a way that they do not affect the real-time capabilities of the underlying CPS.

SmaCCS – Smart Camera Cloud Services.

CamInSens –   Verteilte vernetzte Kamerasysteme zur in situ-Erkennung Personen-induzierter Gefahrensituationen.

Q-Trajectories – 3D-Überwachung von öffentlichen Räumen mit einem Smart-Camera Netz.

OC-Trust – Vertrauenswürdigkeit von Organic Computing Systemen.