Enhancing multivariate process monitoring in soybean meal production using robust statistical quality control techniques

Abstract

Statistical process monitoring of many quality variables independently can lead to confounding results. A production process with two or more correlated quality features, require the use of Multivariate Statistical Process Control and Multivariate Capability Analysis. This study employed Multivariate Statistical Quality Control techniques to analyze the soybean meal production at Hule and Sons Nigeria Limited. Quality characteristic of interest were percentage of crude residual oil in the meal after extraction using solvent extraction method; percentage moisture content; free fatty acid (FFA); amount of phosphorus and the flash point. The study investigates the stability and capability of the multivariate process using advanced statistical process control techniques. Hoteling’s T² square control chart, applied on the transformed data, indicated statistical control with no out-of-control signals detected as all the points fall below the upper control limit of 13.19. In contrast, Robust principal component analysis (ROBPCA) identified one observation exceeding the 95% control limit which suggests the presence of an assignable variation which the conventional method may overlook. Furthermore, the orthogonal distance (OD) chart revealed two samples outside the 95% control threshold, although all the OD values were almost all zeros, indicating potential collinearity or numerical issues in orthogonal projections. The process capability indices were  0.3864; ;   =0.0086 and , signifying a substantial deviation from the desired performance standard. These results highlight the limitations of relying solely on traditional control charts and emphasize the importance of incorporating robust multivariate techniques for more sensitive and reliable process monitoring in high-dimensional data.