How Multivariate Design of Experiments and Feature Extraction Strategy Can Support the Understanding of Time Domain Nuclear Magnetic Resonance Signals from Complex Elastomeric Systems

Abstract Time domain nuclear magnetic resonance (TD-NMR) is a nondestructive analytical technique increasingly used for the characterization of rubber products. Its growing adoption is due to the nature of the acquired signal, which consists of a time-dependent decay strongly correlated with the polymeric mobility of the analyzed materials. Consequently, any factor influencing polymeric mobility also affects the TD-NMR signal, making it essential to understand the impact of these factors for routine application of the technique. To this aim, a multivariate design of experiment (MDoE) approach was employed to evaluate the effects of three factors: (i) sample temperature during TD-NMR analysis at three levels (−1, 0, +1); (ii) the relative amount of three polymers in the sample at two levels (−1, +1); and (iii) the vulcanization state of the sample at two levels (−1, +1). A feature extraction strategy was applied to the resulting data set to derive a single parameter capable of retaining the information content of the full signal. Several parameters were investigated and compared. The most promising was identified as the time required for the signal to reach a specific intensity, representing an innovative and user-friendly index. The combined MDoE and feature extraction approach enables the development of response surfaces describing how the index, and thus the TD-NMR signal, varies with the investigated factors. This methodology provides valuable insights that may support a more informed and effective use of TD-NMR technique.