Enhancing Multi-Hole Pressure Probe Data Processing in Turbine Cascade Experiments Using Structural Risk Minimization Principle

Multi-hole pressure probes are crucial for turbomachinery flow measurements, yet conventional data processing methods often lack generalization for complex flows. This study introduces an innovative approach by integrating machine learning techniques with the structural risk minimization (SRM) principle, significantly enhancing the generalization capability of regression models. A comprehensive framework has been developed, combining SRM-based machine learning regression methods, such as ridge regression and kernel ridge regression, with hyperparameter optimization and S-fold cross-validation, to ensure robust model selection and accuracy. Validated using the McCormick function and applied to VKI-RG transonic turbine cascade measurements, the SRM-based methods demonstrated superior performance over traditional empirical risk minimization approaches, with lower error ratios and higher R2 values. Novel insights from SHAP analysis revealed subtle but significant differences in aerodynamic parameters, including a 0.63122° discrepancy in exit flow angle predictions, guiding the probe design and calibration strategies. This study presents a holistic workflow for improving multi-hole pressure probe measurements under high-subsonic conditions, representing a meaningful enhancement over traditional empirical methods and providing valuable references for practical applications.