Experimental exploration of rheological behavior of polyethylene glycol-carbon dot nanofluid: Introducing a robust artificial intelligence paradigm optimized with unscented Kalman filter technique

• Published in: Journal of molecular liquids 2022-07, Vol.358, p.119198, Article 119198
• Main Authors: Shahsavar, Amin, Amin Mirzaei, Mohamad, Shaham, Aidin, Jamei, Mehdi, Karbasi, Masoud, Seifikar, Fatemeh, Azizian, Saeid
• Format: Article
• Language: English
• Subjects: Artificial neural network; Carbon dot nanofluid; Polyethylene glycol; Response surface methodology; Unscented Kalman filter; Viscosity
• ISSN: 0167-7322; 1873-3166
• DOI: 10.1016/j.molliq.2022.119198

•Rheological behaviour of polyethylene glycol-carbon dot nanofluid is examined.•Effect of shear rate, nanoparticle concentration and temperature are assessed.•A predictive model is developed for estimating the viscosity of the nanofluid.•The nanofluid shows a non-Newtonian behavior.•The UKF-ANN scheme was adopted for accurate estimation of the nanofluid viscosity. In the present study, the polyethylene glycol 200 (PEG200)-based nanofluid containing carbon dot (CD) nanoparticles was synthesized, and its rheological behavior at different temperatures and nanoparticle concentrations (φ) was investigated. The values considered for φ were 0%, 1% and 3% and 7% the values considered for temperature were 20, 30, 40, 50 and 60 °C. It was observed that the PEG200 has a Newtonian behavior, and the nanofluid has a non-Newtonian behavior which is amplified with increasing temperature. Also, a decreasing and increasing trend of viscosity was observed with temperature and φ. As another novelty of this research, a robust novel artificial neural network (ANN) model integrated with an unscented Kalman filter (UKF-ANN) was presented for accurate estimation of the viscosity of the PEG-CD nanofluid based on φ, temperature, and shear rate as the input features. Besides, two efficient data-driven approaches, including classical perceptron ANN (MLP) and response surface methodology (RSM) were developed to examine and evaluate the robustness of UKF-ANN model. The statistical and infographic assessment indicated that the UKF-ANN outperformed the MLP and RSM, respectively.