Model-free predictive control (MFPC) strategies have shown significant potential for achieving robust and highperformance operation in permanent magnet synchronous motor (PMSM) drives without relying on detailed system models. This paper presents a novel MFPC approach, termed the Local Dynamic Estimation Predictive Controller (LDEPC), which integrates a flexible real-time estimator with a refined local dynamics framework to capture unknown system behaviors. Two variants are proposed: the Zero-Order LDEPC, optimized for simplicity and ease of implementation, and the First-Order LDEPC, designed to improve disturbance rejection and capture nonlinear effects more accurately. The framework balances robustness, prediction accuracy, and computational efficiency, enabling effective control under parameter variations, rapid reference changes, and external disturbances. The effectiveness of LDEPC is experimentally validated on a PMSM test bench across multiple scenarios, including step responses, regenerative braking, parameter mismatch, disturbance rejection, and dynamic reference tracking. Comparisons with PI control, an MFPC using an ultra-local model and extended state observer (MFPCC-ESO), and an observer-enhanced MFPC with a generalized proportional-integral observer (MFPCC-GPIO) show that LDEPC consistently delivers robust, reliable, and precise control, underscoring its strong potential for practical deployment in demanding applications.