Validation Principles of Agricultural Machine Multibody Dynamics Models

Abstract

Multibody dynamic (MBD) models are required to be accurate if they are to be used with confidence. To determine the overall accuracy of an agricultural machine MBD model, the model needs to be validated through proper experimental design principles and error analysis. At a high-level, the main methodology concepts for the experimental design involve categorizing dynamic subsystems, creating a data acquisition plan, determining and completing experimental testing, and pre-processing of physical and virtual time series datasets. This paper provides a methodology and example validation process for an MBD model of an agricultural self-propelled sprayer. As part of the experimental testing, a tip test revealed that the vertical center of gravity of the MBD model resulted in a 2.8% error from the physical machine. Furthermore, to summarize simulation model performance against physical data, three error quantification metrics for time-series data comparison are proposed: (1) correlation coefficient, (2) two-sample t-test and two-sample F-test, and (3) standard deviation of absolute error. For determining whether there is an acceptable level of agreement, this study proposes that the correlation coefficient should be greater than 0.7, the two-sample t-test and two-sample F-test should both pass (but with special consideration of the resulting confidence intervals if they fail), and that the standard deviation of absolute error should be less than 25% of the standard deviation of physical time series data. These error metrics, together with their defined sets of agreement criteria, resulted in detecting where the self-propelled sprayer system was experiencing validation issues and eventually the overall practical validation of the MBD model.