A Hybrid System with Multivariate Data Validation and Case Base Reasoning for an Efficient and Realistic Product Formulation

Abstract

This applied research paper presents a novel hybrid system, which provides a systematic approach for an efficient and realistic case retrieval, retention and testing of product formulations. The underlying idea is to build a case library of practical and viable product formulations with consistent quality patterns, flexible process attributes and constituent proportions. To avoid the storage of non-representative and unrealistic cases within the case library, a strict multivariate validation method has been imposed on the system. The input formulation, whether it be a single suggestion on product formulation as a query, an optimized case or a collection of tests, is validated against the most similar formulation cluster in the case library determined through the Principal Component Similarity factor and Mahalanobis distance. T ² and Q-statistics as multivariate data validation methods are employed to determine whether the input formulations match the most similar cluster of datasets in the case library. The synergistic use of univariate control charts and the graphical plot of variable relations between the input formulation and the most similar case provide information on variables, which cause a mismatch. If the value of the culprit variable cannot be rectified to match the dataset in the case library, the new input formulation can only be retained after an empirical validation in the main manufacturing area.