Defining Multivariate Raw Materials Specifications in Industry 4.0

The development of specification regions for raw materials is crucial to ensure the desired quality of the final product. Despite that, nowadays raw materials specifications are usually defined in an arbitrary way according to subjective points of view, based mostly on past experience rather than using a quantitative description of their impact on Critical Quality Attributes (CQAs) of the final product. Besides, in many cases, univariate specifications on each property are used assuming that these properties are independent from one another. But, when univariate specifications are used to select raw materials having correlated properties as it is usually the case, significant amounts of material may be misclassified.

Hence, we propose a methodology, based on the inversion of the Partial Least Squares Regression model, for defining multivariate raw materials specifications linked to CQAs where prediction uncertainty is considered. This is significant because, thus, such specifications provide assurance of quality, and the user can predict if the quality of the final product for a new raw material batch would be achievable without having to manufacture the product with such batch. Besides, this approach can be used with historical data (i.e. daily production data not coming from any experimental design), being very attractive for the Industry 4.0 environment.