Abstract: Robust design is an effective cost methodology to improve product quality by reducing the variation effects of input variables. Mean square error (MSE) is usually regarded as the most appropriate standard for robust optimization design process in the target being best, but with MSE standard the process variation and product quality fluctuation resulted from the noise factors could not be analyzed. The rationality and deficiency of MSE standard and the important about process variance are analyzed. Confidence region of process variance is given by response model. Then an optimizing model of robust parameter design is constructed with MSE as target function and process variance and mean bias as constrictions. The simulation example successfully illustrates the developed model’s advantage. Not only can the process capability achieve six sigma level, but the value of process variance lies in the con-striction of minimization process variance. When minimizing MSE, the process capability is higher than the developed model, but the value of process variance is beyond the constriction. So the solution would can not be guaranteed the process robustness and unexpected variation behavior would occur in practice. The conflict between bias and variance in MSE standard is effectively solved with the restriction of process variance. A robust optimized strategy is set up.

Key words: Confidence region of process variance, Mean square error, Robust design, Six sigma