Validation of the mathematical model of upper body biomechanics using neural networks

Abstract

The paper presents artificial neural networks (ANNs) as a tool for validating the mathematical model. The neural network architecture must learn the relationship between the inputs (which are the parameters of the biomechanical mathematical model, such as height and body mass) and the desired outputs (i.e. the perturbing forces of the biomechanical model elements. We used part of the perturbing force values of the muscle groups responsible for the spine, neck, and head movement, determined using the mathematical model and the C++ application, to train the neural network. We used the remaining data to validate the neural network. The neural network architecture was created using the Easy NN application. After training the network, we concluded that the subject's height has the most significant impact on generating muscle force and is also the most sensitive parameter. The muscle force values of the data used for validation are almost equal to those determined using the mathematical model. Therefore, we can conclude that the mathematical model is correct, and the neural network can make predictions for various subject dimensions, even if their values are not within the range of values for which we trained the network. Keywords: mathematical model; neural networks; human body