Digital Engineering


Digital Engineering research at Webis deals with the development and application of AI technology for solving engineering problems. Our research contributes to simulation data mining, optimum configuration algorithms, smart design support, model-based as well as heuristic diagnosis algorithms, and the advancement of applied numerical analysis in general. Patents have been granted for the integration of model formation and numerical analysis algorithms.

In simulation data mining (Matilda ~ Mining Artificially Generated Data) we are dealing with the generation of knowledge and decision rules from large numbers of simulations. We observe that virtual experimenting by simulation, the basis of today's engineering processes, has become both a blessing and a curse: each aspect of a large system that can be modeled will be simulated, likewise, each parameter that can be varied will be sampled within its domain. The underlying (design) search space is of exponential size and leads to an enormous quantity of simulation data, which is useless for the human analyst (engineer, designer, architect, researcher) if she is not provided with smart technology for data interpretation.

For the automatic synthesis of technical systems we have presented formal models of technical system design, design graph grammars to generate structure models, and machine learning approaches to learn optimum design decisions. We have also compared human design strategies to computer-based heuristic search. Our projects cover basic research funded by the DFG as well as close cooperations with leading industry partners. Part of our interactive design technology has been awarded with scientific and commercial prizes.

In the field of computer-supported diagnosis and fault analysis we have developed approaches to effectively analyze fault behavior in tightly coupled and feedback systems. Basic idea is to break feedback loops by various simulations in all expected failure modes, and to compile a database at the knowledge-level from which, in turn, heuristic rules are learned via data mining. For the application of this approach to automotive settings patents have been granted.

[awards: FluidSIM1, FluidSIM2, ArtDeco]


Students: David Wiesner, Katja Müller, Peter Hirsch, Jens Opolka, Tom Paschke, and Michael Völske.