Reliable and adaptive grasping in robotic systems remains challenging due to the lack of real-time structural feedback and multi-directional stress awareness. To address this, we propose a novel dexterous manipulation system that integrates distributed optical fiber sensing and fuzzy logic-based adaptive control. A strain rosette structure, formed by embedding fiber Bragg gratings (FBGs) in robotic fingers, enables high-resolution two-dimensional (2D) stress field reconstruction with a sensitivity of \(-0.2091\) pm/N and linearity of 0.99. Crucially, instead of relying on conventional single-axis force, the system employs real-time maximum principal stress as a feedback signal to ensure structural safety and enable adaptive grasp force regulation. The fuzzy controller dynamically adjusts control gains to accommodate objects with varying stiffness and damping characteristics. Experimental results demonstrate that this approach reduces steady-state force error to within 6.5 kPa and improves response speed by 18% compared to conventional PID control. This work establishes a closed-loop framework that tightly couples structural integrity assessment and intelligent control, offering a promising solution for safe and efficient robotic grasping in dynamic environments.