The optimization for multidisciplinary engineering systems is highly complicated, which involves the decomposing of a system into several individual disciplinary subsystems for obtaining optimal solutions. Managing the coupling between subsystems remains a great challenge for global optimization as the existing methods involve inefficient iterative solving processes and thus have higher time cost. Some strategies such as discipline reorder, coupling suspension and coupling ignoring can to some extent reduce the execution cost. However, there are still some deficiencies for these approaches such as uniform handling of the couplings, complete decoupling and heavy burden of system optimizer. To overcome the above drawbacks, a serialization-based partial decoupling approach is proposed in this study, which consists of three main steps. First, different disciplines are clustered into some subsystems by analyzing the interdisciplinary sensitivities. Then, for each subsystem, a serialization process is proposed to ensure no coupling loops exist and the subsystem can be solved with no iteration, which can reduce the time cost for solving the disciplinary problem to a large degree. Finally, a local optimization model is constructed for each subsystem to maintain the scale of the global optimizer and ensure mutual independence and parallel processing. The proposed three-layer framework ensures the feasibility of solving for each subsystem and improves the efficiency of optimization execution. Several experiments have been conducted to demonstrate the effectiveness and feasibility of the proposed approach.