A Robust Collaborative Optimization Method Under Multidisciplinary Uncertainty∗

Multidisciplinary design optimization (MDO) is a useful technique on complex product design in recent years. Collaborative optimization (CO) is an effective MDO methods based decomposition which is for deterministic optimization. However, many uncertainties exist in product design such as model error and design variables error. And the propagation of uncertainties in multidisciplinary is more complicated than in a single disciplinary because of the coupled systems. Therefore, robust design has become more important in engineering systems, and its research and applications have extended to multidisciplinary design environment from primary single disciplinary. To make reliable decisions, some researchers have studied several useful methods on multidisciplinary design optimization under uncertainty environment. In this paper, a new robust collaborative optimization (RCO) method is proposed based on system uncertainty analysis (SUA) method. First, given the probabilistic distribution of model error and design variables, the mean and variance of system output is calculated by the SUA method. Then using implicit uncertainty propagation (IUP) method, we get the uncertain estimation of auxiliary design variables that is introduced in CO method. In the following, we embed both SUA and IUP methods into CO method framework, and put the estimation of variables which is output from SUA and IUP methods into CO calculation flow, then the optimization calculation process will not stop until system become convergent. Finally, we realize a new robust collaborative optimization method. Compared to the existing RCO method, our method’s advantage is the probabilistic presentation of uncertainty objective functions and constraints instead of the variation presentation in the exiting RCO method, so we can know more information about the system performance that is influenced by uncertainty design parameters and variables. Then we can make more reliable decisions in designing the engineering systems according to the probability distributed of system objective function.