Optimization-Driven Synthetic Data Generation for Surrogate Models with Cross-Domain Application: A Concept Study with Compressor Blade Fillets

Kurzfassung

This study presents a surrogate modeling framework for the dimensioning of compressor blade components demonstrated with an application for the design of compressor blade fillets. The compressor blade, a critical component, demands a highly specialized design approach, where the fillet is of particular interest due to its impact on the part's aerodynamic and structural performance. Our research introduces a fast, robust, and reliable method that employs machine learning techniques to enhance the compressor blade fillet design process. The method extends upon a validated prototype process for compressor blisks, as outlined in previous work. Significant advancements to the prototype are achieved in two primary aspects of this process. First, enhancements to the optimization algorithm. For this task we use a specialized open-source framework for multidisciplinary design, analysis, and optimization to balance the conflicting demands of structural durability and minimizing fillet size. The second major enhancement pertains to the automation of data pipelines, covering the data generation, processing, training and validation.