Achieving robustness in engineering design is crucial for systems such as gas-bearing supported turbocompressors used in heat pumps and fuel cells, where precision affects efficiency and stability. Traditional design methods often optimize for nominal conditions rather than real-world variabilities, largely due to the extensive computational resources required for accurate modeling. This thesis introduces an innovative, automated, AI-based design framework for all-at-once optimization that addresses these challenges more effectively.

The study integrates ensemble artificial neural networks (EANNs) with the Non-Sorted Genetic Algorithm-III (NSGA-III) for comprehensive multidisciplinary design optimization (MDO). This integration accurately models interdependencies among subsystems like bearings, rotors, and impellers and was validated against high-fidelity models, achieving significant computational speed enhancements and real-time simulation capabilities.

The EANNs improved computational speeds by four orders of magnitude, integrating robustness metrics like hypervolume and signal-to-noise ratio directly into the optimization objectives. This approach achieved manufacturing tolerances as large as ±8 micrometers for herringbone grooved journal bearings, enhancing the feasible region for rotor stability by 78% against manufacturing deviations.

Experiments validated the surrogate models' accuracy, extending the EANNs' application to spiral groove thrust bearings and axial dynamics, culminating in a holistic, robust optimization of compressor systems ranging from 1kW to 8kW. The designs showed efficiencies exceeding 82% in higher-power applications and pinpointed an optimal robustness at 3kW.

The thesis led to the ARMID (Automated Robust Modular Integrated Design) framework, integrating surrogate modeling, optimization and real-time simulation tools with an automated 3D CAD generation tool (ParaturboCAD), producing over a thousand CAD models available in STEP and STL formats. This framework enhances the accessibility, robustness, and efficiency of high-precision systems, significantly advancing the field of robust multidisciplinary design optimization.