A Study Of Global Robust Parameter Design For Complex Relationship Processes

Robust Parameter Design (RPD) is a main method to increase the stable of manufactures and decrease the fluctuation of products’quality. As technology advances, there are too many manufacturing processes with complex relationship in numerous industries. The processes have multi-extremes, and high-order nonlinear relationship between their influential factors and response output. Considering on the experiment design’s ways and the restriction of modeling’s form, the existing robust parameter design can only optimize local parameter, but is not for the complex processes. How quality characteristic’s expectation and variance can realize effective global modeling in the feasible region, and then achieve global parameter optimization in complex relationship processes? The question became the key points for quality improvement of manufacturing. Basing on the global robust, the paper explores the parameter optimization for complex relationship processes, and the main contents are as follows:The paper constructs the single response modeling basing on support vector machine. It adopts the single response modeling method to study the influence of controllable factors’ variation and noise factors’fluctuations on the response output’s fluctuations. It combines the factors’ joint probability density distributions, and expresses the influence through the form of multiple integral. With the excellent generalization performance of support vector machine, it establishes continuous variable global modeling of expectation and variance for the complex relationship processes. The single response modeling method is used to fit regression function, then the paper takes some points to make contrast analysis between fitting value and the actual value and the dual response’s fitting value, and it is shown that the proposed method has good fitting and prediction performance.The paper studies the global parameter optimization with the modeling that established. It proposes block strategy of the feasible region for researching optimization in the "larger the better" questions and "small the better" questions, and puts forward using uniform design to construct initial population in connection with genetic algorithm and chooses the corner and center points as the initial points aiming at sequential quadratic programming. After that, it uses the improved algorithm to solve the optimal value of the nonlinear problem.The paper conducts the simulation and empirical study by using proposed modeling and optimization method. The single response modeling is applied to fit the approximate models for the inductor-resistor series circuit and product nuts experiment, the optimization strategy is used to optimize parameter of the regression modeling, then the best combination scheme of factors is selected by analysis. The validation results display that the proposed modeling and optimization method are effectiveness and practicability in the study of the robust parameter design for the complex relationship processes.The paper proposes the thinking, the realization methods, the technical flow about the global parameter optimization for the complex relationship processes. The findings expand the research field of robust parameter design, have current relevance and high values on reducing the processes’ fluctuations and improving the production’s quality.