This paper presents the development and validation of a fiber-optic sensing–based inclinometer system for distributed shape and deformation monitoring of slender structural elements. Distributed deformation monitoring is essential for structural assessment and operational safety, yet conventional approaches often provide only local information and require complex instrumentation. To address these limitations, an 18 m long glass-fiber-reinforced polymer (GFRP) rod is instrumented with three embedded single-mode optical fibers, evenly spaced by 120° around the cross-section. At eight discrete cross-sections along the rod, Fiber Bragg Gratings (FBGs) are integrated into each fiber, providing three strain measurements per section and enabling reconstruction of in-plane bending behavior and geometric deformation states along the rod length. To establish a robust measurement system and generate accurate, traceable reference data, a controlled deformation environment is implemented. Well-defined deformation states are imposed using an automated actuation concept, allowing complex but repeatable bending shapes to be produced under consistent boundary conditions. A digital representation of the setup is used to prescribe target geometries at the measurement sections and to derive the corresponding actuation settings required to reproduce these shapes. The resulting configurations are applied to the physical system, enabling deformation states to be generated in a consistent and verifiable manner. During each deformation state, the corresponding FBG strain responses are recorded, resulting in a high-quality dataset linking deformation configurations and strain measurements. The dataset is used to train a neural network that maps measured strain information to deformation-related parameters. To assess the influence of data representation on predictive performance and robustness, the same network architecture is trained using four different data preparation strategies. The models are evaluated using standard regression metrics and systematically compared to identify robust and informative input representations. The presented approach supports reliable sensor validation and provides practical guidance for machine-learning-based deformation prediction toward real-world deployment of distributed fiber-optic shape sensing.