Study On Multi-Response Robust Design Based On RSM

In this dissertation, the methodologies for Robust Parameter Design (RPD) of the products or processes with multiple characteristics are developed. The objective is to develop methods for mult-iresponse RPD when noise factors and fluctuations of design parameters are considered. The research includes the following aspects.First, it relates to the improvement of the traditional desirability function method. Desirability function method is the most popular approach for multi-response optimization, however, the variation, correlation and the uncertainty of parameter estimation of the responses are ignored in this approach. Thus an improved desirability function method is presented. The implementation and effectiveness of the proposed method are illustrated by an example from the literature. The result shows, the proposed method which is robust to prediction quality and variant of responses yields better results than traditional desirability function method.Secondly, it presents a method to multi-response RPD when noise factors are presented. The mean and variances of the responses are estimated, and the variance estimation is improved to avoid the possible nonpositive definiteness of estimated variance. Then the multivariate loss function method is improved to consider the influence of noise factors, and a new desirability function method is proposed to take into account the mean squared error of the responses. Both the mean value and variance models are incorporated into the two methods, in which the variance combines the variance due to the noise factors with the variance due to predictions, which lead to an unbiased estimator of the combined variance. Finally, an example is given where the two methodologies are applied to the data, the results show that the solutions obtained by proposed methods are robust to both noise factors and parameter estimation uncertainty.Thirdly, it addresses the multi-response RPD issue when both the fluctuations of design parameters and influences of noise factors are considered. When the design parameters follow normal distribution, the mean and variance estimated models are presented, and the new desirability function method which combines variances due to noise factors and the variation of the controllable factors, the sensitivisity of the responses to the fluctuation of controllable factors is given. The effectiveness of the proposed method is illustrated by an example from the literature. The result shows, compared with results from the literature, the robustness to the fluctuation of controllable factors is improved. When the design parameters drift or degraded over operation time, the quality losses are calculated, and the multi-response RPD methodology are explored when the design parameters are reset based on simple block reseeting policy, modified block resetting policy and age resetting policy.At last, it extends the concurrent optimization of parameter design and tolerance design to the case of multiple responses when the design parameters follow normal distribution. A total cost model is proposed by balancing multivariate losses incurred by parameter design and increased costs caused by tightening the tolerances of design parameter. And the variations of design parameter and economies of the process are considered by the proposed model. The implementation and advantages of the proposed method are illustrated by an example. The result indicates, compared with conventional parameter design, the variances of the responses and the total costs are all reduced dramatically. Furthermore, the integrated optimization of parameter design, tolerance design and maintenance is explored for products or processes with multiple characteristics when the design parameters degraded. The total cost function which consists of quality losses, manufacturing cost and maintenance is minimized to simultaneously obtain the optimal settings and tolerances of design parameter, as well maintenance policy.