Ensuring stability is important requirement when performing movements in human and robotic mechanisms. Our goal is to develop a simple and portable measurement system for assessing stability parameters. Wearable magneto-inertial measurement units are used to evaluate stability parameters Center of Mass (CoM) and Zero Moment Point (ZMP) which are compared to the Center of Pressure (CoP) reference parameter from the force plate.
At the beginning of the master's thesis we devote a few words to stability of mechanisms, follows presentation of existing measurement systems for stability observation and description of workflow including goals of research study. On the sequence is given an explanation of biomechanical parameters of the body and a description of the process of estimating the orientation of segments with the kinematic model according to the Denavit-Hartenberg principle. The main part of this chapter is devoted to the presentation of CoP, CoM and ZMP parameters and their interconnections between motion. The methodology and the measurement protocol are described in the third chapter, where we present the measuring system consisting of force plate and magneto-inertial measuring units. Chapter then describes experimental measurements conducted with three test subjects. The fourth chapter is dedicated to the presentation of a multivariate linear regression model and a neural network algorithm, used for assessing the CoP position by using wearable magneto-inertial measurement sensors.
The results are divided into five parts. First, the graphs show trajectory of CoP, CoM and ZMP in walking, stepping in place, swinging body around ankles, and randomly stepping over force plate. While evaluating ZMP, results shown no statistical significance; on the other hand, just the opposite is valid for CoM trajectory, which well coincides with CoP. In the second set we compare the CoP on the graph courses which evaluated from the multivariate linear regression model with reference CoP trajectory from force plate. In tables we show the performance parameters with correlation coefficients, the mean absolute error, and root square mean error. We also evaluate the performance of the neural network algorithm. Then, in the fourth set of results, we find a better performance of the neural network algorithm in the CoP assessment by comparing the average values of absolute percentage errors and average correlation coefficients. The last set of results gives information on the performance of the CoP evaluation with respect to the different combinations of sensors.