Cloud-Based Generative-AI Supporting Preliminary Engineering Design

Many engineering solutions require specialized know-how and a deep understanding of the physics mechanisms that underpin their design and operation. Concurrently, there is a growing need for enhanced design space exploration capabilities overcoming the limitations of parametric models, enabling the assessment of innovative design concepts through free-form geometry modelling approaches. This session outlines the collaborative work between Rolls-Royce and Databricks, with a primary focus on optimizing Conditional Generative Adversarial Network (cGAN) training processes. Leveraging Databricks and acquired know-how resulted in a significant reduction in runtime, approximately by a factor of 30, achieved through distributed computing for parallel hyper-parameter tuning. The integration of MLflow ensures transparency and reproducibility. Furthermore, the implementation of Unity Catalog establishes a crucial governance framework for compliance-centric industries, including aerospace.