The spring constants of primary suspensions for a railway vehicle are optimized by a robust design process, in which the response surface models(RSMs) of their dynamic responses are constructed via the design of experiment(DOE). The robust design process requires an intensive computation to evaluate exactly a probabilistic feasibility for the robustness of dynamic responses with their probabilistic variances for the railway vehicle. In order to overcome the computational process, the process capability index $C_{pk}$ is introduced which enables not only to show the mean value and the scattering of the product quality to a certain extent, but also to normalize the objective functions irrespective of various different dimensions. This robust design, consequently, becomes to optimize the $C_{pk}$ subjected to constraints, i.e. 2, satisfying six sigma. The proposed method shows not only an improvement of some $C_{pk}$ violating the constraints obtained by the conventional optimization, but also a significant decrease of the variance of the $C_{pk}$.