Achieving full multi-objective optimisation in MAENAD. To achieve the optimisation objectives in the MAENAD project, there are three main steps that must be fulfilled: It must be possible to define the design space. This is primarily a language issue and means ensuring that EAST-ADL and its variability mechanisms can be used to define optimisation-based variability as well as normal product line variability. Although the language definitions are largely beyond the scope of this document, in practical terms these mechanisms must also be harmonised with HiP-HOPS and any plugins, so that tool support exists for them. It must be possible to evaluate the design candidates. This will require a close link with both existing and developing analysis tools. HiP-HOPS already has a direct connection with its safety analysis elements but to allow other forms of analysis, e.g. timing, it will likely require some substantial modifications to the optimisation engine, particularly if the engine is to use a modelling tool such as Papyrus as its base. The multi-objective optimisation heuristics must allow for an efficient exploration of the design space. In practice, the demands of accomplishing the MAENAD objectives are likely to require extensions, modifications, and tweaks to the optimisation algorithm implemented in HiP-HOPS. The first step is the variability support. If it is not possible to define the design space, then no optimisation can take place. The central principle here is substitutability: the language has to allow the designer to specify alternative versions of model elements while ensuring that the alternatives are sufficiently constrained such that they can readily be substituted for one another by the optimisation algorithm. If swapping one alternative for another results in an invalid model, then they are not substitutable. Of particular difficulty here is ensuring a consistent interface; some variants may have more ports or connections than other variants, and so it has to be possible to ensure that those connections remain valid regardless of which variant is chosen. For example, if a variant with more connections to other components is chosen, then failure propagation should be possible along those connections. This leads on to the second step - that of evaluation of design variants. The notion of substitutability is encapsulated in practice by the process of variability resolution wherein a model containing variability is resolved to produce a new model in which all the variability has been fixed to...
