On Sampling-Times-Independent Identification of Relaxation Time and Frequency Spectra Models of Viscoelastic Materials Using Stress Relaxation Experiment Data.

Viscoelastic relaxation time and frequency spectra are useful for describing, analyzing, comparing, and improving the mechanical properties of materials. The spectra are typically obtained using the stress or oscillatory shear measurements. Over the last 80 years, dozens of mathematical models and algorithms were proposed to identify relaxation spectra models using different analytical and numerical tools. Some models and identification algorithms are intended for specific materials, while others are general and can be applied for an arbitrary rheological material. The identified relaxation spectrum model always depends on the identification method applied and on the specific measurements used in the identification process. The stress relaxation experiment data consist of the sampling times used in the experiment and the noise-corrupted relaxation modulus measurements. The aim of this paper is to build a model of the spectrum that asymptotically does not depend on the sampling times used in the experiment as the number of measurements tends to infinity. Broad model classes, determined by a finite series of various basis functions, are assumed for the relaxation spectra approximation. Both orthogonal series expansions based on the Legendre, Laguerre, and Chebyshev functions and non-orthogonal basis functions, like power exponential and modified Bessel functions of the second kind, are considered. It is proved that, even when the true spectrum description is entirely unfamiliar, the approximate sampling-times-independent spectra optimal models can be determined using modulus measurements for appropriately randomly selected sampling times. The recovered spectra models are strongly consistent estimates of the desirable models corresponding to the relaxation modulus models, being optimal for the deterministic integral weighted square error. A complete identification algorithm leading to the relaxation spectra models is presented that requires solving a sequence of weighted least-squares relaxation modulus approximation problems and a random selection of the sampling times. The problems of relaxation spectra identification are ill-posed; solution stability is ensured by applying Tikhonov regularization. Stochastic convergence analysis is conducted and the convergence with an exponential rate is demonstrated. Simulation studies are presented for the Kohlrausch-Williams-Watts spectrum with short relaxation times, the uni- and double-mode Gauss-like spectra with intermediate relaxation times, and the Baumgaertel-Schausberger-Winter spectrum with long relaxation times. Models using spectrum expansions on different basis series are applied. These studies have shown that sampling times randomization provides the sequence of the optimal spectra models that asymptotically converge to sampling-times-independent models. The noise robustness of the identified model was shown both by analytical analysis and numerical studies.