Statistical analysis of identification of linear viscoelastic models

Viscoelastic materials (VEMs) have gained increasing popularity for their ability to dampen vibrations in various structural applications in recent years. The mechanical characteristics of VEMs can be effectively described using constitutive models featuring both integer and fractional derivatives. This study examines the mechanical behavior of VEMs using fractional Zener models with four, five, and six parameters, as well as the generalized Maxwell model with 16 parameters, which relies on integer derivatives. To accomplish this, the study formulates an optimization problem with the aim of minimizing an error function defined by the quadratic relative distance between theoretical model responses and experimental data. Solving the optimization problem involves the use of a hybrid optimization technique, which combines genetic algorithms and non-linear programming. After obtaining optimal designs for each viscoelastic model, qualitative assessments demonstrate that all analytical models provide satisfactory fits to the experimental data. Subsequently, a statistical analysis employing Akaike’s Information Criterion is conducted to identify the models that best describe the mechanical behavior of the analyzed VEMs. In this quantitative evaluation encompassing all viscoelastic models, it is noted that the generalized Maxwell model with 16 terms produces a lower relative error and statistically outperforms the fractional Zener models only in a global analysis. However, in a temperature-by-temperature analysis, the GMM16 proves to be inferior to all fractional models. Furthermore, when focusing solely on the fractional models, the five-parameter Fractional Zener Model exhibits the best statistical fit to the experimental data.