Reynolds-averaged Navier–Stokes (RANS) turbulence models lose predictive accuracy in separated-flow regimes. To address this limitation, a closed-form correction based on Gaussian radial basis function (RBF) [1] was introduced into the Spalart–Allmaras turbulence model. Unlike black-box neural net-work models (NN), the RBF provides an analytical relationship between the turbulence model correction and local flow variables. The RBF parameters were optimized using the Field Inversion and Machine Learning-Direct [2] approach by minimizing the discrepancy between RANS predictions and experimental data. Training was performed using a multi-case dataset that only consists of single-element airfoil (SEA) configurations. As shown in Fig. 1, the augmented model predicts the lift coefficient with significantly smaller deviations than the baseline model under stall conditions. However, when the augmented model is applied to the DLR-F15 multi-element airfoil (Fig. 2), it gives poor predictions, reflecting its training on SEA configurations. In contrast, a classical field inversion, which was performed on the DLR-F15 case, captures flow separation at the trailing edge of the flap with high accuracy. Based on these findings, a new RBF model and NN model were trained. While the new RBF model gives only a very slight change towards the desired results, the new NN model tends to yield results similar to the baseline solution, however, yielding a larger separated region and lower suction peaks on the three airfoil elements. Consequently, this work will focus on developing a new training dataset that includes high-lift configurations to extend the applicability of the model to industrially relevant aerodynamic flows.