Abstract: In view of that the feasible region is possibly inexistent during the collaborative optimization at system level, the genetic algorithm (GA) is introduced in combination with strengthening the constraint conditions step by step to develop a new GA-based collaborative optimization algorithm. In this algorithm the individual in feasibility in population is computed with the optimal values resulting from subsystems, then whether the solution to an individual is feasible depends on the infeasibility and its threshold. A method to adjust the threshold is proposed via cyclic iteration steps to ensure that the optimization at system level will go towards decreasing the unsatisfiability of constraint on consistency equality, thus achieving the goal to enhance the interdisciplinary consistency. Taking the design of a speed reducer as example, the performance of the optimization algorithm is proved excellent.