Paper Number
PO21 

Session
Poster Session

Title
Bayesian credibility analysis of viscosity-temperature models relevant to Redox Flow Battery (RFB) working fluids

Presentation Date and Time
October 12, 2022 (Wednesday) 6:30

Track / Room
Poster Session / Riverwalk A

Authors (Click on name to view author profile)

1. Gupta, Shatakshi (University of Illinois at Urbana-Champaign, Mechanical Science and Engineering)
2. Ewoldt, Randy H. (University of Illinois at Urbana-Champaign, Mechanical Science and Engineering)

Author and Affiliation Lines (in printed abstract book)
Shatakshi Gupta and Randy H. Ewoldt
Mechanical Science and Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801

Speaker / Presenter 
Gupta, Shatakshi

Keywords
experimental methods; computational methods; rheometry techniques

Text of Abstract
A statistical credibility analysis has been carried out to compare more than 15 viscosity-temperature relationships for fluids relevant to Redox Flow Batteries (RFBs). This includes solvents like Acetonitrile (ACN), and ionic liquids such as 1-Butyl-1-methylpyrrolidinium bis(trifluoromethylsulfonyl)imide (Pyr14TFSI) for which the viscosities were experimentally measured at RFB-relevant working conditions such as 20 °C – 60 °C across several shear rates. Newtonian behavior, which is typical for small molecule fluids considered in this study, has been observed up to shear rates of 39,000 s-1 for ACN and 3,800 s-1 for Pyr14TFSI using an internal flow pressure-driven slit rheometer (m-VROC, Rheosense). The experimentally obtained temperature-dependent viscosities were fitted using the existing viscosity-temperature models whose mathematical forms can be classified into Arrhenius-like, Power Law, algebraic, and other empirical forms. The goodness of the fit has been examined through the Residual Sum of Squares weighted by the measurement uncertainty of the datapoints. Though most models show good values of the regression coefficient (R2 > 0.9) in the tested temperature range, it is necessary to evaluate the model credibility for comparison. Fit credibility of the models was assessed using the Bayesian Information Criterion (BIC) to tradeoff the closeness of the fit with over-parameterizing. For PyR14TFSI, we observe that the Vogel-Fulcher-Tammann (VFT) model, which has widely been used in literature for ILs shows a good R2 value but lower fit credibility (relatively higher BIC value) when compared to other models. Although Arrhenius forms show slightly poorer fit credibility than some power-law models, they have a physical basis in statistical mechanical considerations that provide molecular-level insights. To account for this difference in the credibility of the physical and empirical models, we discuss the quantification process of a pre-factor (“prior”) in the “likelihood” term of the BIC formulation.