The tolerance distribution model of quality control points oriented to assembly accuracy was established to characterize the nonlinear coupling relationship among the different kinds of quality control points and reveal accumulation law of assembly error. The chaos theory and two-stage variation strategy were adopted to improve multiple objective particle swarm optimization (MOPSO). Before the assembly process, the initial distribution results of quality control points were calculated. The redistribution mechanism was implemented when upstream quality control points were instantiated. Then the tolerance of downstream quality control points would be redistributed to provide decision support for the instantiation of subsequent quality control points. The online optimization and former feed-forward control of assembly quality were achieved. The online optimization of gasket-selection and tighten lock nut in the active gear axle assembly process was given to verify the feasibility and effectiveness of the proposed method.