Wrinkled surfaces, characterized by strictly periodic microstructures, can serve as templates to provide localized, repetitive microdeformations and open routes to controlled deformation properties important for strain engineering. Low-pressure plasma treatment with various process gases tunes the localized curvature conditions of the microstructures, resulting in adjusted bending radii ranging from 20 to 3200 nm. A key challenge in achieving appropriate bending conditions is predicting the required process conditions. We demonstrate that machine learning techniques solve this prediction problem and allow precise bending adjustments. Therefore, we designed an artificial neural network (ANN) that directly maps the process parameters to the resulting wavelength (λ) and amplitude (A). By coupling this model with a bisection method, we solved the inverse design problem, allowing the derivation of possible combinations of process parameters to achieve a specific λ or A. This transforms the fabrication process from trial-and-error to a precisely controlled engineering workflow. Thus, our finding opens new perspectives for the tailored fabrication of wrinkled surfaces with repetitive, controlled bending in manifold applications, such as strain engineering of 2D-materials for optical and catalytically active surfaces, or structure-dependent localized charge accumulation, e.g., for enhanced performance of triboelectric-based sensors.