An Innovative Method for Predicting Parametric Dependent-Pressure Drops of Meandering Magnetorheological Valves Using Deep Neural Networks

Prabhakara, Hafizh Arsa; Imaduddin, Fitrian; Bahiuddin, Irfan; As’arry, Azizan; Ismail, Andhi Akhmad; Mazlan, Saiful Amri

doi:10.24200/sci.2025.65007.9243

An Innovative Method for Predicting Parametric Dependent-Pressure Drops of Meandering Magnetorheological Valves Using Deep Neural Networks

Articles in Press

Document Type : Research Article

Authors

Hafizh Arsa Prabhakara ¹

Fitrian Imaduddin ²

Irfan Bahiuddin ³

Azizan As’arry ⁴

Andhi Akhmad Ismail ¹

Saiful Amri Mazlan ⁵

¹ Department of Mechanical Engineering, Vocational College, Universitas Gadjah Mada, Yogyakarta, Indonesia

² - Mechanical Engineering Department, Faculty of Engineering, Islamic University of Madinah, Medina, Saudi Arabia - Mechanical Engineering Department, Faculty of Engineering, Universitas Sebelas Maret, Surakarta, Indonesia

³ Universitas Gadjah Mada

⁴ Department of Mechanical and Manufacturing Engineering, Faculty of Engineering, Universiti Putra Malaysia, Serdang, Selangor, Malaysia

⁵ Malaysia-Japan International Institute of Technology, Universiti Teknologi Malaysia Kuala Lumpur, Malaysia

10.24200/sci.2025.65007.9243

Abstract

This paper introduces a novel approach for predicting magnetorheological (MR) valve pressure drop using deep neural networks. Rather than regressing the pressure drop directly, the proposed methodology predicts magnetic flux densities across different zones of the valve and then uses these predictions to compute the MR fluid’s yield stress and the valve’s pressure drop. The proposed approach can be further deployed for optimization purposes and provides insight into the magnetic field distribution within the MR valve as a function of design parameters. The approach leverages finite element simulations encompassing 125 variations of geometric parameters (gap sizes) and control parameters (electrical currents). A multilayer neural network architecture is tuned by testing 72 configurations of activation functions, numbers of hidden nodes, and layers, with model selection based on mean squared error and R^2. The final model demonstrates high fidelity with R^2 more than 0.98 for both training and testing. By capturing how magnetic flux density varies as a function of design and control parameters, the proposed framework facilitates efficient optimization and design of MR valves without exhaustive simulation. These results underscore the method’s ability to capture the complex dynamics governing MR valve pressure drop and provide valuable insights for valve sizing and performance prediction.

Keywords

design

deep neural networks

semi-active devices

magnetorheological valves

pressure drop

Subjects