Abstract
Keywords: Genetic algorithm, Optimisation, Dependability
A shift towards architecturally distributed, networked and functionally integrated system architectures is becoming apparent in many engineering system designs in the automotive industry. New and emerging designs for car control systems are characterised by sharing of information and hardware resources which means that large numbers of reconfiguration options are available at design but also at runtime due to the use of shared processors and communication channels.
When the functions of a system can be delivered with a number of different architectural configurations designers are faced with hard optimisation problems. The dependability of individual configurations of the design can be determined using safety assessment and verification technologies such as fault tree analysis and model checking. However, satisfying dependability requirements with an optimal use of resources and minimal costs requires additional technological support in the form of a global optimisation process. Where it is possible to fulfil all the dependability requirements, within economical and technical bounds, the architecture that has minimal costs is the optimisation goal. If all dependability requirements can not be met with acceptable costs then the problem changes to finding architectures that achieve optimal trade-off between dependability and cost.
The substitution of single components either with functionally equivalent but more reliable components or with fault tolerant schemes that include replicated components in redundant configurations are mechanisms that designers have for altering the safety and reliability characteristics of a system. This approach can be effective when combined with measures to ensure component (both hardware and software) diversity in redundant configurations, minimising the potential for common cause failures.