Constraint-Based Evolutionary Learning Approach to the Non-normal ProcessPerformance Evaluation

Performance of industrial products is very important for an industry. Conventional methods for performance analysis consider a normality assumption and limited to low dimensional data. Different manufacturing processes very often have products with quality characteristics that do not follow normal distribution. In such cases fitting a known non-normal distribution to these quality characteristics would lead to erroneous results while assessing the performance of products. In this paper, we propose a novel method for non-normal multivariate process performance analysis. However, optimal estimation of the parameters of non-normal multivariate distribution is a crucial problem in process performance analysis. We have proposed a Constraint-based Evolutionary Algorithm (EA) approach for optimal estimation of the parameters of non-normal multivariate process. Furthermore, a geometric distance based method has been employed to reduce higher dimensionality of data to lower dimension. The efficacy of the proposed method is assessed by using the proportion of nonconformance (PNC) criterion to summarize the performance of EA approach. The experimental results from constraint-based EA have been compared to those obtained using steepest descent and simulated annealing (SA) approaches.