A neural network applied to estimate process capability of non-normal processes

It is always crucial to estimate process capability index (PCI) when the quality characteristic does not follow normal distribution, however skewed distributions come about in many processes. The classical method to estimate process capability is not applicable for non-normal processes. In the existing methods for non-normal processes, probability density function (pdf) of the process or an estimate of it is required. Estimating pdf of the process is a hard work and resulted PCI by estimated pdf may be far from real value of it. In this paper an artificial neural network is proposed to estimate PCI for right skewed distributions without appeal to pdf of the process. The proposed neural network estimates PCI using skewness, kurtosis and upper specification limit as input variables. Performance of proposed method is validated by simulation study for different non-normal distributions. Finally, a case study using the actual data from a manufacturing process is presented.