The accurate energy consumption modeling of robot arms is crucial for energy conservation and optimization. Traditionally, parametric dynamics model is used for estimating the robot arm’s torque and power consumptions. However, accuracy of parametric dynamics model relies on the dynamic parameters such as inertia, center of mass, friction etc. In practice, these parameters are hard to estimate and optimization usually require extensive identification process. Similarly, a data-driven approach requires large amount of observation data, are computationally inefficient and suffer from occasional inaccuracies. In this study, a hybrid learning approach combining both parametric model and data-driven method is proposed for energy consumption modelling of industrial robot arm in static pose. Parametric model with approximate estimates of the robot arm dynamics is used to simulate the energy consumption and an artificial neural network is used to learn the error between the simulation and the observed results. Halton-sequence sampling is used for collecting the training data with joint angles as input and energy consumption as output. The effectiveness of proposed model is verified using experimental data and the proposed approach achieves significantly lower mean squared error and higher R-square value than the parametric and the data-driven models while only using approximate dynamic parameters for base dynamics model.
Analytical modeling, Artificial neural networks, Data-driven modeling, Energy, Hybrid learning, Machine Learning, Robots
This work is licensed under a Creative Commons Attribution 4.0 International License.
Creative Commons CC BY 4.0