Towards Bayesin Co-Design for Mission Specific Design of Dynamical Systems

Highly dynamical systems are an integral part of tomorrow’s society, ranging from robots, over floating wind turbines to drones. The conventional approach to the development of these systems relies on the V-diagram, in which we move from concept to product over an iterative loop of design, implementation and testing. Up until recently a sequential approach has been pursued during the design phase, where first the design is optimized statically after which its functionality is enhanced by optimizing its control trajectory. This impedes finding systems with concurrent optimal design and trajectory. To address this, co-design methods have appeared that do this simultaneously. Up until now, co-design has only been applied to low-fidelity models, that are cheap to evaluate but typically lack the ability to correctly represent reality and lead to many testing cycles. In this work we push model-based system design further, pursuing a ‘first time right’ paradigm, through the inclusion of high-fidelity models (such as for example computational fluid dynamics). To account for the computational cost of these models, we introduce Bayesian optimization, leading to Bayesian co-design. To validate the new methodology the model-based design of a drone was performed with the objective of performing inspection services for wind turbine parks. The technological progress that was realized does not restrict itself to drones, but is applicable to a broader scope of highly dynamical systems. As such it contributing to breaking through the current barriers of several fields to accelerate the transition towards a better future.