Quaternion Valued Neural Network for Modeling Vibrating Electromechanical Structures to Reduce Physical Vibrating Hazard in the Work Environment

In this contribution, the physical risk factor problem related to work tools due to which the operator is subjected to continuous low-frequency stresses is addressed in terms of modeling and prediction. The problem is approached from the experimental point of view by means of the realization of a vibrating platform that under appropriate stresses makes it possible to reproduce real operating conditions. By means of a quaternion-based hypercomplex neural network, experimental data are processed in view of an n-step prediction to predict the dynamics of the system through measurements of accelerations and angular velocities obtained by means of a simple inertial platform. The results obtained are satisfactory and hold promise for real-world applications in which the number of sensors used is minimized and, above all, they can be located in protected areas in order to ensure their functionality and resistance to failure.