A Study on Data-Driven Surrogate Modeling for Friction Force in a Hydraulic Actuator Using Deep Neural Networks

EasyChair Preprint 13496

Abstract

Accurate control and efficient operation of hydraulically driven systems rely on the precise identification of the friction force within hydraulic actuators. Predicting the friction force is challenging due to its inherent nonlinearities and complex physical nature. This study introduces a data-driven approach using deep neural networks (DNN) to predict nonlinear friction forces. Various modeling techniques for hydraulic actuators exist, with the lumped fluid theory being a widely used method due to its efficiency and accuracy. The LuGre friction model is commonly used to describe friction forces, incorporating variables such as bristle deformation and tangential velocity. The DNN was trained with data from a uniaxial hydraulic actuator and utilized inputs including pressures, actuator length, velocity, and acceleration. The DNN demonstrated its predictive performance in a four-bar mechanism simulation, effectively replacing the mathematical friction model.

Keyphrases: Deep Neural Networks, friction force, hydraulic actuator

@booklet{EasyChair:13496,
  author    = {Seongji Han and Grzegorz Orzechowski and Jin-Gyun Kim and Aki Mikkola},
  title     = {A Study on Data-Driven Surrogate Modeling for Friction Force in a Hydraulic Actuator Using Deep Neural Networks},
  howpublished = {EasyChair Preprint 13496},
  year      = {EasyChair, 2024}}