The rhythmic chewing movement pattern is dynamically reshaped to adapt to a variable chewing environment. The variance affects the wear performance of dental prostheses. This study was aimed to generate these variable rhythmic chewing movements for dental testing equipment. A six-axis parallel chewing robot DUT-2 was adopted as the dental testing equipment. Four variances were extracted from the rhythmic movement, including period, offset, amplitude, and mode. The relevant movement cases, including gradually accelerating movement, gradually increasing movement, gradually shrinking movement, and bilateral movement, were designed. Then, a central pattern generator (CPG) model based on morphed phase oscillators was proposed. According to the coupling feature of the rhythmic movements, the specific modulation method of the CPG model was provided for these movements. The simulated incisor trajectory was outputted by importing the driving amplitudes from the CPG model to the virtual prototype of the chewing robot. The bite force (considering two-body and three-body contacts) was analyzed by writing the driving amplitudes into the motion controller of the chewing robot’s physical prototype. The relative errors of offset and amplitude in the z-direction were 4.14% and 0.74%, respectively. The transition was smooth around the turning point during the gradually increasing movement and bilateral movement. For two-body contact, the average relative error and bias of the maximum bite force were 4.15% and 1.08%, respectively. The food involvement decreased the accuracies to 13.18% and 2.50%, respectively. The CPG model supplies a bionic and explicit approach for generating the variable rhythmic chewing movements. The variable movements related to chewing preferences and food properties could be replicated. Besides, the high repeatability of the maximum bite force is beneficial for running the repetitive wear tests. Finally, the CPG model makes it possible to study the influence of the variance on wear performance.