A Novel Spatial Transformer Framework for Compressor Map Adaptation

Abstract Accurate gas turbine modeling is crucial for performance evaluation, control optimization, and fault diagnosis, especially under off-design conditions or component degradation. Traditional thermodynamic models rely on complete compressor and turbine maps, but these are often proprietary, outdated, or incomplete, limiting their applicability. Hybrid approaches that integrate physical models with data-driven techniques offer a promising alternative but face challenges due to limited real-world data, difficulty in capturing nonlinear relationships in degraded component maps, and inadequacies in modeling spatial relationships in compressor maps. To address these challenges, this study proposes a novel framework that generates degraded compressor maps, samples operating curves from these maps, and trains a neural network to predict performance under varying conditions. A stage-stacking method with embedded decay factors creates diverse degraded maps for training, simulatingawide range of operating scenarios. A Spatial Transformer Network is then employed to model the affine transformation between generalized performance maps and degraded conditions, enabling adaptive adjustments to reflect real-world scenarios. The results of the numerical experiments demonstrate that the proposed framework effectively captures the relationship between operating lines and affine transformations, enhancing the adaptability of compressor maps. This approach shows significant potential for improving gas turbine monitoring and control. Future work will focus on improving training data with more accurate degradation models, exploring advanced neural network architectures, and validating the framework with real-world operational data.