This paper presents a modular, multi-disciplinary and multi-fidelity surrogate-based optimization framework for rotor blade design, aiming to reduce computational cost while preserving ”useful accuracy” to reliably guide design decisions. The workflow relies on a 3D blade parametrization and integrates aerodynamic and mechanical evaluations within an online, high-dimensional and highly constrained optimization loop [1]. A key focus is the assessment of blade/casing contact robustness through a direct comparison of two modeling levels: (i) simplified clearance consumption criteria and (ii) non-linear contact simulations. The objective is to quantify to what extent simplified criteria can efficiently steer the more expensive non-linear simulations when used as constraints to an industrial optimization process. Results show that both approaches locate minima in similar regions of the design space, indicating that simplified criteria provide effective guidance. However, some regions with low non-linear responses are not ranked as favorably by the simplified criteria, which may occasionally drive the optimization away from attractive configurations. For aerodynamic modeling, a multi-fidelity surrogate strategy combines low-cost 2.5D simulations with higher-fidelity 3D CFD at two operating conditions. The 2.5D evaluations are leveraged to enhance the surrogate quality while limiting the required number of expensive 3D simulations, enabling efficient exploration of the design space while enforcing aerodynamic constraints. On the mechanical surrogate side, advanced Non-Intrusive Proper Orthogonal Decomposition (NI-POD) models are compared with Tuned Radial Basis Functions (TRBFs) for the prediction of criticality fields. NI-POD consistently improves prediction accuracy of derived quantities, but with a noticeable increase in computational cost. The framework is demonstrated on the design of an industrial low-pressure compressor rotor blade, maximizing efficiency at cruise conditions under aerodynamic and mechanical constraints. Overall, the study highlights how combining multi-fidelity aerodynamics with simplified contact robustness criteria can efficiently guide aero-mechanical design, providing a practical balance between precision and computational cost.