Mandibular movement is essential for mastication and speech, driven and controlled by a complex neuro-musculoskeletal system involving the temporomandibular joint, ligaments, muscles, and neural control. As a high-dimensional, redundant, and multimodal system, its coordinated control dynamics remain unclear, which significantly impacts clinical diagnosis, and rehabilitation. This study proposes a multibody dynamics simulation framework that considers neuromuscular rhythm control and muscle synergy. Using jaw kinematics and surface electromyography as modeling inputs, a multibody dynamics model of the mandibular musculoskeletal system is established, and a forward-inverse coupling framework is used to obtain muscle activation patterns. The obtained time-dependent data are then decomposed into a weight matrix and temporal patterns using non-negative matrix factorization, and the trajectory sensitivity of each muscle is subsequently evaluated via parameter perturbation. According to the results, muscle bundles are classified into four functional units (high/low synergy × high/low sensitivity). Using Bayesian optimization, the muscle parameters of each group are fine‑tuned, reducing the mandibular trajectory error from 3.2 mm to 2.1 mm. To simulate rhythmic mastication patterns, a central pattern generator (CPG)‑based control approach is introduced. A three‑neuron Matsuoka oscillator network is constructed, and each neuron unit corresponds to the synergistic muscle control for jaw opening, lateral movement, and jaw closing, respectively. The network parameters are optimized using a genetic algorithm to reproduce mastication kinematics in trajectory shape, rhythm, and muscle synergy. The proposed framework successfully generates a periodic mastication task without using measured mandibular kinematics as model inputs. This study establishes a multi‑level computational framework that successfully integrates computational musculoskeletal biomechanics, neuromuscular control, and nonlinear optimization algorithms. It not only provides an efficient tool for motor control but also offers new insights into functional partitioning and synergistic biomechanics of the neuromuscular system. The proposed framework lays a theoretical foundation for designing multimodal control strategies in intelligent rehabilitation devices.