Comparative study on the machine learning-based techniques for magnetorheological elastomer dynamic properties prediction

• Main Authors: Saharuddin, Kasma Diana, Ariff, Mohd Hatta Mohammed, Bahiuddin, Irfan, Mazlan, Saiful Amri, Nazmi, Nurhazimah, Ubaidillah, Ubaidillah
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; Back propagation networks; Comparative studies; Dynamic characteristics; Elastomers; Field strength; Gaussian process; Loss modulus; Machine learning; Magnetic fields; Magnetic properties; Neural networks; Nonlinear control; Support vector machines; Viscoelasticity
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0229967

Magnetorheological elastomers (MRE) have gained popularity due to their ability to control viscoelastic properties by varying the strength of the magnetic field. Due to the obvious nonlinear and complex behavior of MRE, machine learning approaches were used to predict the MRE viscoelastic properties, which are storage and loss modulus. In comparison to the traditional viscoelastic model, which is complex in mathematical derivation, machine learning method easily identifies trends and patterns by mapping the input-output relationship. It can also handle nonlinear problems by training on data. Support vector regression (SVR), Gaussian process regression (GPR), Backpropagation neural network (BP-ANN), and Extreme learning machine (ELM) were introduced and compared to simulate the field-dependent viscoelastic behavior of MRE with frequency and magnetic field strength as model input. As a result, the ELM model produced the highest accuracy, with more than 98 percent accuracy on model generalization capability. Therefore, this demonstrates that machine learning can replace traditional modelling approaches and serve as a basis for material and device development.