Tagungsband - DAGA 2024
Beitrag
Machine Learning-based Analysis of a Physics-Informed Data Set of a Viscoelastic Damped Membrane
Zusammenfassung:
Viscoelastic damping plays an important role in the sound produced by percussion instruments. Research suggests that it is frequency dependent and leads to non-exponential decay, spectral sidebands and mode coupling. Nonetheless, a methodic analysis of viscoelastic damping in musical instruments is a challenging task since many of the available data sets do not account for a systematic modification, and therefore annotation, of the parameters that define its behavior. To overcome this problem, this work presents a sound data set generated by a Finite-Difference Time Domain viscoelastic model of a membrane. Four parameters can be modified within the model: the viscoelastically damped frequency, the strength of the damping, the rate of decay of the memory effect, and the length of the viscoelastic buffer. To create each sound the parameter space is modified systematically, generating a physics-informed data set. Subsequently, the amplitude peaks of the resulting frequency spectra are analyzed by a Self-Organizing Map (SOM). Preliminar results discussing the non-linearities of the viscoelastic damping caused by different parameter settings are presented.