Two novel methods to analyze the combined effect of multiple raw-materials and processing conditions on the product's final attributes: JRPLS and TPLS

• Cross raw material interactions are important to consider to data mine a process.
• Raw material's physical properties are a key source of information for data mining.
• JRPLS is designed to accommodate the data structure from a multi-material scenario.
• TPLS method encompasses process, material's properties, mixture ratios and quality.
• JRPLS and TPLS parameters provide valuable interpretability for process improvement.

The challenge in the design of a new product or the troubleshooting of a manufacturing operation is to understand how the materials, the composition of the materials and the processing conditions affect the properties of the product. Two new methods (JRPLS and TPLS) are proposed to overcome the gaps in the current mixture modeling methods to simultaneously analyze data from multiple materials, their mixing ratios and processing conditions as to how they affect the quality of a product. Unlike the prior alternatives, the methods proposed here consider explicitly the physical properties of the raw materials, without the need of a linear weighted average estimation. A NIPALS-like algorithm is proposed to estimate the parameters of both models, and analytical expressions are derived for the JRPLS demonstrating that it reduces to the LPLS in the scenario of one material. The methods are illustrated with three industrial case studies involving the analysis of the manufacturing process for three different drug products. In all cases the JRPLS and TPLS methods enable the identification of key relationships between materials, process and products.