Model-Based Multi-objective Safety Optimization

It is well-known that in many safety critical applications safety goals are antagonistic to other design goals or even antagonistic to each other. This is a big challenge for the system designers who have to find the best compromises between different goals. In this paper, we show how model-based safety analysis can be combined with multi-objective optimization to balance a safety critical system wrt. different goals. In general the presented approach may be combined with almost any type of (quantitative) safety analysis technique. For additional goal functions, both analytic and black-box functions are possible, derivative information about the functions is not necessary. As an example, we use our quantitative model-based safety analysis in combination with analytical functions describing different other design goals. The result of the approach is a set of best compromises of possible system variants. Technically, the approach relies on genetic algorithms for the optimization. To improve efficiency and scalability to complex systems, elaborate estimation models based on artificial neural networks are used which speed up convergence. The whole approach is illustrated and evaluated on a real world case study from the railroad domain.