Model-Based Systems & Qualitative Reasoning
Group of the Technical University of Munich
Work on model-based systems and qualitative reasoning at the Computer Science Institute (Chair IX) of the Technical University of Munich started when Peter Struss joined the faculty in 1992. He had been working in AI since 1978 and on research in that particular area since 1981 at Siemens Corp. Research and Development. A basic principle of our work is to always carry out and evaluate basic research in the context of concrete case studies, usually provided by some industrial problem solving task.
Systems for automated fault analysis, diagnosis, testing, and repair of technical systems
The group has developed tools and applications dedicated to supporting diagnosis of real physical system. This includes the automated generation of diagnostic hypotheses, proposed tests and repair actions. This is done based on a component-oriented behavior model of the respective device. Models are also used to support activities during design, such as diagnosability analysis, sensor placement, and Failure Mode Effect Analysis (FMEA), a method used for the identification of potential error types and their effects on the examined object and to classify them with respect to criticality or diagnosability. In the framework of several European and national projects, we have been collaborating with other academic partners and companies, in particular with car industries. Ref. Struss, P.: Modellbasierte Systeme und Qualitative Modellierung. Erschienen in: Görz, G., Rollinger, C., Schneeberger, J. (Hrsg.): Handbuch der Künstlichen Intelligenz, 3. Auflage, Oldenbourg Verlag, München, 2000.
Model-based behavior prediction, situation assessment and therapy proposal for natural and technical processes, in particular ecological systems
In complex ecosystems, it is often a knowledge intensive task to infer the causes of undesirable developments. One has to know possible phenomena and their preconditions. Assessing the situation of an ecosystem from a limited number of observations can be defined as a challenging diagnosis task. Additionally, one wants to reason about possible cures or symptom treatments in order to influence the ecosystem in a direction that is in accordance with some specified goals. We have developed theoretical foundations and systems for such an extended diagnosis task and apply the results to water treatment problems. Project AGUA
Automated and qualitative modeling
Many tasks, for example in diagnosis or during early stages of design, have to be performed without precise numerical information, and sometimes, our knowledge about functional interdependencies underlying a class of systems (e.g. ecosystems) is partial and qualitative. We develop and apply methods and systems for representing and reasoning with incomplete knowledge about physical systems. Qualitative modeling aims at representing all possible behaviors consistent with the given partial knowledge. In order to be applicable within industrial application systems, modeling itself has to be supported and automated. This includes the automated composition of models from model fragments (compositional modeling? and the automated generation of qualitative models from numerical models.