Characterization and modeling of a new magnetorheological damper with meandering type valve using neuro-fuzzy

• Main Authors: Imaduddin, F., Mazlan, S. A., Ubaidillah, Ubaidillah, Idris, M. H., Bahiuddin, I.
• Format: Article
• Language: English
• Published: Elsevier B.V. 2017
• Subjects: T Technology (General)
• Online Access: http://eprints.utm.my/id/eprint/74854/1/IrfanBahiuddin2017CharacterizationandModelingofanewMagnetorheologicalDamper.pdf; http://eprints.utm.my/id/eprint/74854/; https://www.scopus.com/inward/record.uri?eid=2-s2.0-85028804228&doi=10.1016%2fj.jksus.2017.08.012&partnerID=40&md5=18c73bd813b650f1bcb2da5060b44058

This paper presents the characterization and hysteresis modeling of magnetorheological (MR) damper with meandering type valve. The meandering type MR valve, which employs the combination of multiple annular and radial flow passages, has been introduced as the new type of high performance MR valve with higher achievable pressure drop and controllable performance range than similar counterparts in its class. Since the performance of a damper is highly determined by the valve performance, the utilization of the meandering type MR valve in an MR damper could potentially improve the damper performance. The damping force characterization of the MR damper is conducted by measuring the damping force as a response to the variety of harmonic excitations. The hysteresis behavior of the damper is identified by plotting the damping force relationship to the excitation displacement and velocity. For the hysteresis modeling purpose, some parts of the data are taken as the training data source for the optimization parameters in the neuro-fuzzy model. The performance of the trained neuro-fuzzy model is assessed by validating the model output with the remaining measurement data and benchmarking the results with the output of the parametric hysteresis model. The validation results show that the neuro-fuzzy model is demonstrating good agreement with the measurement results indicated by the average relative error of only around 7%. The model also shows robustness with no tendency of growing error when the input values are changed.