Application of artificial intelligence to magnetite-based magnetorheological fluids

• Published in: Journal of industrial and engineering chemistry (Seoul, Korea) 2021, 100(0), , pp.399-409
• Main Authors: Saberi, Hossein, Esmaeilnezhad, Ehsan, Choi, Hyoung Jin
• Format: Article
• Language: English
• Subjects: Artificial neural network; Fuzzy logic; Magnetite; Magnetorheology; 화학공학
• ISSN: 1226-086X; 1876-794X
• DOI: 10.1016/j.jiec.2021.04.047

[Display omitted] •MR fluids become attractive as smart materials.•Shear stresses of MR fluids with four input parameters are examined using artificial neural networks.•Multilayer perceptron neural network provides the best response.•Equation based on network weights and biases is presented for predicting shear stress of magnetite-based MR fluids. Magnetorheological (MR) fluids are intelligent fluids that change their state under a magnetic field and can be extensively applied in several industries. In this study, a model was presented to predict the MR behavioral trend of magnetite-based MR fluids using deep neural networks. The MR data of nine samples with several magnetite nanoparticle concentrations and different silicone oil viscosities were used for network construction and testing; the aforementioned data were obtained under several magnetic field strengths. Seven samples were used for network training/testing within the training interval and two samples were applied for evaluating the network accuracy outside the network training interval. Several networks, such as the multi-layer perceptron (MLP), radial basis function, and adaptive neuro-fuzzy inference system, were employed, and the results were analyzed. The accuracy parameters (R2 and RMSE) of the MLP network for the training data (0.99625 and 0.00867) and test data (0.99130 and 0.01621), as well as a comparison between the predicted and laboratory-measured results of the two samples that had not been used in the modeling step, demonstrated the exceptional performance of the proposed method and an equation that was derived for predicting the shear stress. The latter equation enables researchers to achieve their needs without performing time-and cost-consuming MR tests in the laboratory.