Hybrid Reliability Analysis Method for Electromagnetic Design Problems with Non-Gaussian Probabilistic Parameters

This paper proposes an efficient and stable reliability analysis method for reliability-based electromagnetic design problems with non-normal probability distributions of input parameters. The reliability analysis strongly depends on distribution types of random variables since nonlinear transformations between an original random space and a standard normal random space cause additional nonlinearity into the reliability assessment of probabilistic constraint functions. That can lead to numerical inaccuracy and instability in the reliability-based design process, or may fail to have a solution to the probabilistic constraint assessment. To overcome these difficulties, a hybrid mean-value method is introduced to seeking a most probable failure point in the performance measure approach, which is one of the first-order reliability analysis methods. The proposed method is tested with a mathematical model and a loudspeaker design, of which random variables are assumed to follow five different probability distributions case by case.