Data Mining and Knowledge Discovery Approach for Manufacturing in Process Data Optimization

Abstract

The in-process data offers a rich source for discovering new process knowledge. This is particularly important in manufacturing processes such as casting process where the number of process variables are large (~30–50) and process observations are small (~50–60). A multivariate data analysis technique such as principal component analysis (PCA) is used for discovering knowledge from foundry in-process data. The correlations among factors for a given response are discovered by projecting the data on a reduced dimensional space defined by the principal components. The correlations are discovered among both categorical and continuous data. A new methodology has been introduced that uses scores and loadings in the PCA to define optimal and avoid tolerance limits for factors. Interactions among factors are also considered. The developed approach allows process engineers to adjust system parameters as it discovers factor tolerance limits that contribute most to the overall variance. This information is used to suggest corresponding optimal and avoid limits that would result in the reduction of variance. The workings of the algorithm are demonstrated on a foundry case study.