Machine Learning Prediction of TiO₂-Coating Wettability Tuned via UV Exposure

Abstract

Surfaces with extreme wettability (too low, superhydrophobic; too high, superhydrophilic) have attracted considerable attention over the past two decades. Titanium dioxide (TiO2) has been one of the most popular components for generating superhydrophobic/hydrophilic coatings. Combining TiO2 with ethanol and a commercial fluoroacrylic copolymer dispersion, known as PMC, can produce coatings with water contact angles approaching 170°. Another property of interest for this specific TiO2 formulation is its photocatalytic behavior, which causes the contact angle of water to be gradually reduced with rising timed exposure to UV light. While this formulation has been employed in many studies, there exists no quantitative guidance to determine or tune the contact angle (and thus wettability) with the composition of the coating and UV exposure time. In this article, machine learning models are employed to predict the required UV exposure time for any specified TiO2/PMC coating composition to attain a certain wettability (UV-reduced contact angle). For that purpose, eight different coating compositions were applied to glass slides and exposed to UV light for different time intervals. The collected contact-angle data was supplied to different regression models to designate the best method to predict the required UV exposure time for a prespecified wettability. Two types of machine learning models were used: (1) parametric and (2) nonparametric. The results showed a nonlinear behavior between the coating formulation and its contact angle attained after timed UV exposure. Nonparametric methods showed high accuracy and stability with general regression neural network (GRNN) performing best with an accuracy of 0.971, 0.977, and 0.933 on the test, train, and unseen data set, respectively. The present study not only provides quantitative guidance for producing coatings of specified wettability, but also presents a generalized methodology that could be employed for other functional coatings in technological applications requiring precise fluid/surface interactions.

Predicting Sliding Angles on Random Pit-Distributed Textures Using Probabilistic Neural Networks

“…The accuracy of the neural network prediction model is illustrated by counting the number of specimens with the match predicted and measured sliding angles as a percentage of the number of specimens. The prediction accuracy of the model was 90.2%, which is close to the results of previous studies. ,, It is worth noting that the PNN model is only suitable for microtexture and cannot accurately predict complex hierarchical structures. The SA of the random pit texture was first predicted by the PNN model.…”

“…The prediction accuracy of the model was 90.2%, which is close to the results of previous studies. 25,26,39 It is worth noting that the PNN model is only suitable for microtexture and cannot accurately predict complex hierarchical structures. The SA of the random pit texture was first predicted by the PNN model.…”

Determination of the heat capacity of cellulosic biosamples employing diverse machine learning approaches

“…The RBFNN designed the approach in three main layers, employing a nonlinear activation function for developing the architecture. The GRNN is a one‐pass learning topology and a memory‐based method that minimizes errors using probability density functions 61 . The radial basis (Equation ) and linear (Equation ) transfer functions are utilized in the hidden and output layers of the RBFNN and GRNN models.…”

Estimating the density of deep eutectic solvents applying supervised machine learning techniques

“…This neuron-based machine learning method is the most widely-used tool as either estimator 65 , 66 or classifier 67 . The working process of the artificial neural network is handled by a combination of linear (LPart) and non-linear (NLPart) operations conducted by the neuron as follows 68 : w , b , and are weight and bias coefficients and activation function, respectively. Although a linear activation function exists, the non-linear, continuous, and differentiable ones often provide artificial neural networks with a better generalization ability 69 .…”