Initial-Parameter-Criterion based Gaussian Mixture Model Monitoring Method for Non-Gaussian Process

Gaussian mixture model (GMM) monitoring method is commonly used for non-Gaussian process monitoring. In GMM monitoring method, the final number of Gaussian components and the statistical distribution parameters are estimated through the improved Figueiredo-Jain (F-J) algorithm in the modeling phase. However, the data dimensionality of modern industry is high, the number of initial Gaussian components is large for the accuracy of the algorithm. Under this condition, the relative small number of samplings makes the denominator in the weight estimation formula of the improved F-J algorithm become zero and the algorithm suspends. To make the algorithm work well in an acceptable modeling time, we deduce the Initial-Parameter-Criterion among initial Gaussian components, data dimension, sampling number and propose the Initial-Parameter-Criterion based GMM (IPC-GMM) monitoring method. The proposed monitoring method is used for TE process monitoring, simulation results reveal the effectiveness and necessity of IPC-GMM monitoring method.