The genetic algorithm and the adaptive mechanism are adopted to tackle the inefficiency of optimization and the convergence difficulty of collaborative optimization (CO). Based on the further analysis of collaborative optimization process, the constraint conditions are converged into part of the optimization function. The system optimization model of CO has been reconstructed according to the adaptive penalty function which is based on the information of different disciplines and the transformation of system-level constraints. Therefore, the global and local search capabilities of optimization algorithm and searching efficiency of CO have been improved. Meanwhile, the difficulty of convergence caused by the internal definition of CO has been resolved. Finally, an example of speed reducer is demonstrated to verify the proposed method, showing that the convergence rate and search efficiency have been improved.