Automated Exploratory Analysis of Spectroscopic Data for Raw Material Identification (RMID) in the Tyre Industry.

Description

The correct identification of raw materials plays a key role in the quality control of the tyre industry production chain. Especially when dealing with a large number of raw materials having different chemical compositions, and with analysis performed by several operators in different plants, it is important to perform raw material identification (RMID) being as more accurate and automated as possible. In this context, near-infrared (NIR) spectroscopy is one of the most suitable techniques, as it is non destructive, user-friendly and timesaving. In the present study, performed in collaboration with Pirelli Tyre S.p.A., FT-NIR spectra (Perkin Elmer, Frontier NIR, 10000 – 4000 cm-1) were acquired for most of the raw materials used for tyre production in two of the main Pirelli Tyre's factories. Exploratory analysis by means of principal component analysis (PCA) was performed on the data with the aim to investigate data structures and groupings within samples. Due to the huge amount of data collected, it was necessary to automate the exploratory analysis through the creation of a dedicated routine in the Matlab ® (Version 2022b, The MathWorks, Inc.) environment. This automation strategy includes incoming controls for each spectrum acquired, with the aim of detecting anomalous data and for the transformation of the measurement scale (from reflectance to pseudo-absorbance). If there are measurement replicates, the average spectrum is then calculated. Supervised classification strategies were then applied on this data, in order to define the class membership of raw material analyzed, in terms of both material specification and supplier. With the aim to apply this strategy for the routine analysis performed in the context of quality assurance, a report including all steps of data import and processing is then generated. A further improvement of this RMID protocol will be the development of predictive models in order to estimate chemical and physical characteristics of the raw material directly during the acceptance phase. Once developed and validated, the proposed strategy will be implemented in a graphical user interface (GUI) created in the Python environment for an immediate application in the industrial production chain; its usage will not require the operator to have previous knowledge about spectroscopy and chemometrics.

Additional details