-
Evaluation of GlassNet for physics-informed machine learning of glass stability and glass-forming ability
Authors:
Sarah I. Allec,
Xiaonan Lu,
Daniel R. Cassar,
Xuan T. Nguyen,
Vinay I. Hegde,
Thiruvillamalai Mahadevan,
Miroslava Peterson,
**cheng Du,
Brian J. Riley,
John D. Vienna,
James E. Saal
Abstract:
Glasses form the basis of many modern applications and also hold great potential for future medical and environmental applications. However, their structural complexity and large composition space make design and optimization challenging for certain applications. Of particular importance for glass processing is an estimate of a given composition's glass-forming ability (GFA). However, there remain…
▽ More
Glasses form the basis of many modern applications and also hold great potential for future medical and environmental applications. However, their structural complexity and large composition space make design and optimization challenging for certain applications. Of particular importance for glass processing is an estimate of a given composition's glass-forming ability (GFA). However, there remain many open questions regarding the physical mechanisms of glass formation, especially in oxide glasses. It is apparent that a proxy for GFA would be highly useful in glass processing and design, but identifying such a surrogate property has proven itself to be difficult. Here, we explore the application of an open-source pre-trained NN model, GlassNet, that can predict the characteristic temperatures necessary to compute glass stability (GS) and assess the feasibility of using these physics-informed ML (PIML)-predicted GS parameters to estimate GFA. In doing so, we track the uncertainties at each step of the computation - from the original ML prediction errors, to the compounding of errors during GS estimation, and finally to the final estimation of GFA. While GlassNet exhibits reasonable accuracy on all individual properties, we observe a large compounding of error in the combination of these individual predictions for the prediction of GS, finding that random forest models offer similar accuracy to GlassNet. We also breakdown the ML performance on different glass families and find that the error in GS prediction is correlated with the error in crystallization peak temperature prediction. Lastly, we utilize this finding to assess the relationship between top-performing GS parameters and GFA for two ternary glass systems: sodium borosilicate and sodium iron phosphate glasses. We conclude that to obtain true ML predictive capability of GFA, significantly more data needs to be collected.
△ Less
Submitted 19 March, 2024; v1 submitted 15 March, 2024;
originally announced March 2024.
-
GlassNet: a multitask deep neural network for predicting many glass properties
Authors:
Daniel R. Cassar
Abstract:
A multitask deep neural network model was trained on more than 218k different glass compositions. This model, called GlassNet, can predict 85 different properties (such as optical, electrical, dielectric, mechanical, and thermal properties, as well as density, viscosity/relaxation, crystallization, surface tension, and liquidus temperature) of glasses and glass-forming liquids of different chemist…
▽ More
A multitask deep neural network model was trained on more than 218k different glass compositions. This model, called GlassNet, can predict 85 different properties (such as optical, electrical, dielectric, mechanical, and thermal properties, as well as density, viscosity/relaxation, crystallization, surface tension, and liquidus temperature) of glasses and glass-forming liquids of different chemistries (such as oxides, chalcogenides, halides, and others). The model and the data used to train it are available in the GlassPy Python module as free and open source software for the community to use and build upon. As a proof of concept, GlassNet was used with the MYEGA viscosity equation to predict the temperature dependence of viscosity and outperformed another general purpose viscosity model available in the literature (ViscNet) on unseen data. An explainable AI algorithm (SHAP) was used to extract knowledge correlating the input (physicochemical information) and output (glass properties) of the model, providing valuable insights for glass manufacturing and design. It is hoped that GlassNet, with its free and open source nature, can be used to enable faster and better computer-aided design of new technological glasses.
△ Less
Submitted 20 November, 2023; v1 submitted 27 March, 2023;
originally announced March 2023.
-
Glass Hardness: Predicting Composition and Load Effects via Symbolic Reasoning-Informed Machine Learning
Authors:
Sajid Mannan,
Mohd Zaki,
Suresh Bishnoi,
Daniel R. Cassar,
Jeanini Jiusti,
Julio Cesar Ferreira Faria,
Johan F. S. Christensen,
Nitya Nand Gosvami,
Morten M. Smedskjaer,
Edgar Dutra Zanotto,
N. M. Anoop Krishnan
Abstract:
Glass hardness varies in a non-linear fashion with the chemical composition and applied load, a phenomenon known as the indentation size effect (ISE), which is challenging to predict quantitatively. Here, using a curated dataset of over approx. 3000 inorganic glasses from the literature comprising the composition, indentation load, and hardness, we develop machine learning (ML) models to predict t…
▽ More
Glass hardness varies in a non-linear fashion with the chemical composition and applied load, a phenomenon known as the indentation size effect (ISE), which is challenging to predict quantitatively. Here, using a curated dataset of over approx. 3000 inorganic glasses from the literature comprising the composition, indentation load, and hardness, we develop machine learning (ML) models to predict the composition and load dependence of Vickers hardness. Interestingly, when tested on new glass compositions unseen during the training, the standard data-driven ML model failed to capture the ISE. To address this gap, we combined an empirical expression (Bernhardt law) to describe the ISE with ML to develop a framework that incorporates the symbolic law representing the domain reasoning in ML, namely Symbolic Reasoning-Informed ML Procedure (SRIMP). We show that the resulting SRIMP outperforms the data-driven ML model in predicting the ISE. Finally, we interpret the SRIMP model to understand the contribution of the glass network formers and modifiers toward composition and load-dependent (ISE) and load-independent hardness. The deconvolution of the hardness into load-dependent and load-independent terms paves the way toward a holistic understanding of composition and ISE in glasses, enabling the accelerated discovery of new glass compositions with targeted hardness.
△ Less
Submitted 19 January, 2023;
originally announced January 2023.
-
Machine learning unveils composition-property relationships in chalcogenide glasses
Authors:
Saulo M. Mastelini,
Daniel R. Cassar,
Edesio Alcobaça,
Tiago Botari,
André C. P. L. F. de Carvalho,
Edgar D. Zanotto
Abstract:
Due to their unique optical and electronic functionalities, chalcogenide glasses are materials of choice for numerous microelectronic and photonic devices. However, to extend the range of compositions and applications, profound knowledge about composition-property relationships is necessary. To this end, we collected a large quantity of composition-property data on chalcogenide glasses from SciGla…
▽ More
Due to their unique optical and electronic functionalities, chalcogenide glasses are materials of choice for numerous microelectronic and photonic devices. However, to extend the range of compositions and applications, profound knowledge about composition-property relationships is necessary. To this end, we collected a large quantity of composition-property data on chalcogenide glasses from SciGlass database regarding glass transition temperature ($T_g$), Young's modulus ($E$), coefficient of thermal expansion (CTE), and refractive index ($n_D$). With these data, we induced predictive models using three machine learning algorithms: Random Forest, K-nearest Neighbors, and Classification and Regression Trees. Finally, the induced models were interpreted by computing the SHAP (SHapley Additive exPlanations) values of the chemical features, which revealed the key elements that significantly impacted the tested properties and quantified their impact. For instance, Ge and Ga increase $T_g$ and $E$ and decrease CTE (three properties that depend on bond strength), whereas Se has the opposite effect. Te, As, Tl, and Sb increase $n_D$ (which strongly depends on polarizability), whereas S, Ge, and P diminish it. Knowledge about the effect of each element on the glass properties is precious for semi-empirical compositional development trials or simulation-driven formulations. The induced models can be used to design novel chalcogenide glasses with required combinations of properties.
△ Less
Submitted 14 June, 2021;
originally announced June 2021.
-
Is the Structural Relaxation of Glasses Controlled by Equilibrium Shear Viscosity?
Authors:
Ricardo F. Lancelotti,
Daniel R. Cassar,
Marcelo Nalin,
Oscar Peitl,
Edgar D. Zanotto
Abstract:
Knowledge of relaxation processes is fundamental in glass science and technology because relaxation is intrinsically related to vitrification, tempering as well as to annealing and sev-eral applications of glasses. However, there are conflicting reports -- summarized here for different glasses -- on whether the structural relaxation time of glass can be calculated using the Maxwell equation, which…
▽ More
Knowledge of relaxation processes is fundamental in glass science and technology because relaxation is intrinsically related to vitrification, tempering as well as to annealing and sev-eral applications of glasses. However, there are conflicting reports -- summarized here for different glasses -- on whether the structural relaxation time of glass can be calculated using the Maxwell equation, which relates relaxation time with shear viscosity and shear modulus. Hence, this study aimed to verify whether these two relaxation times are comparable. The structural relaxation kinetics of a lead metasilicate glass were studied by measuring the re-fractive index variation over time at temperatures between 5 and 25 K below the fictive temperature, which was initially set 5 K below the glass transition temperature. Equilibrium shear viscosity was measured above and below the glass transition range, expanding the current knowledge by one order of magnitude. The Kohlrausch equation described very well the experimental structural relaxation kinetics throughout the investigated temperature range and the Kohlrausch exponent increased with temperature, in agreement with studies on other glasses. The experimental average structural relaxation times were much longer than the values computed from isostructural viscosity, as expected. Still, they were less than one order of magnitude higher than the average relaxation time computed through the Maxwell equation, which relies on equilibrium shear viscosity. Thus, these results demon-strate that the structural relaxation process is not controlled by isostructural viscosity, and that equilibrium shear viscosity only provides a lower boundary for structural relaxation kinetics.
△ Less
Submitted 15 March, 2021; v1 submitted 20 October, 2020;
originally announced October 2020.
-
Predicting and interpreting oxide glass properties by machine learning using large datasets
Authors:
Daniel R. Cassar,
Saulo Martiello Mastelini,
Tiago Botari,
Edesio Alcobaça,
André C. P. L. F. de Carvalho,
Edgar D. Zanotto
Abstract:
With the advent of powerful computer simulation techniques, it is time to move from the widely used knowledge-guided empirical methods to approaches driven by data science, mainly machine learning algorithms. We investigated the predictive performance of three machine learning algorithms for six different glass properties. For such, we used an extensive dataset of about 150,000 oxide glasses, whic…
▽ More
With the advent of powerful computer simulation techniques, it is time to move from the widely used knowledge-guided empirical methods to approaches driven by data science, mainly machine learning algorithms. We investigated the predictive performance of three machine learning algorithms for six different glass properties. For such, we used an extensive dataset of about 150,000 oxide glasses, which was segmented into smaller datasets for each property investigated. Using the decision tree induction, k-nearest neighbors, and random forest algorithms, selected from a previous study of six algorithms, we induced predictive models for glass transition temperature, liquidus temperature, elastic modulus, thermal expansion coefficient, refractive index, and Abbe number. Moreover, each model was induced with default and tuned hyperparameter values. We demonstrate that, apart from the elastic modulus (which had the smallest training dataset), the induced predictive models for the other five properties yield a comparable uncertainty to the usual data spread. However, for glasses with extremely low or high values of these properties, the prediction uncertainty is significantly higher. Finally, as expected, glasses containing chemical elements that are poorly represented in the training set yielded higher prediction errors. The method developed here calls attention to the success and possible pitfalls of machine learning algorithms. The analysis of the SHAP values indicated the key elements that increase or decrease the value of the modeled properties. It also estimated the maximum possible increase or decrease. Insights gained by this analysis can help empirical compositional tuning and computer-aided inverse design of glass formulations.
△ Less
Submitted 24 May, 2021; v1 submitted 1 September, 2020;
originally announced September 2020.
-
Designing optical glasses by machine learning coupled with a genetic algorithm
Authors:
Daniel R. Cassar,
Gisele G. dos Santos,
Edgar D. Zanotto
Abstract:
Engineering new glass compositions have experienced a sturdy tendency to move forward from (educated) trial-and-error to data- and simulation-driven strategies. In this work, we developed a computer program that combines data-driven predictive models (in this case, neural networks) with a genetic algorithm to design glass compositions with desired combinations of properties. First, we induced pred…
▽ More
Engineering new glass compositions have experienced a sturdy tendency to move forward from (educated) trial-and-error to data- and simulation-driven strategies. In this work, we developed a computer program that combines data-driven predictive models (in this case, neural networks) with a genetic algorithm to design glass compositions with desired combinations of properties. First, we induced predictive models for the glass transition temperature ($T_g$) using a dataset of 45,302 compositions with 39 different chemical elements, and for the refractive index ($n_d$) using a dataset of 41,225 compositions with 38 different chemical elements. Then, we searched for relevant glass compositions using a genetic algorithm informed by a design trend of glasses having high $n_d$ (1.7 or more) and low $T_g$ (500 °C or less). Two candidate compositions suggested by the combined algorithms were selected and produced in the laboratory. These compositions are significantly different from those in the datasets used to induce the predictive models, showing that the used method is indeed capable of exploration. Both glasses met the constraints of the work, which supports the proposed framework. Therefore, this new tool can be immediately used for accelerating the design of new glasses. These results are a step** stone in the pathway of machine learning-guided design of novel glasses.
△ Less
Submitted 15 March, 2021; v1 submitted 20 August, 2020;
originally announced August 2020.
-
ViscNet: Neural network for predicting the fragility index and the temperature-dependency of viscosity
Authors:
Daniel R. Cassar
Abstract:
Viscosity ($η$) is one of the most important properties of disordered matter. The temperature-dependence of viscosity is used to adjust process variables for glass-making, from melting to annealing. The aim of this work was to develop a physics-informed machine learning model capable of predicting $η(T)$ of oxide liquids. Instead of predicting the viscosity itself, the NN predicts the parameters o…
▽ More
Viscosity ($η$) is one of the most important properties of disordered matter. The temperature-dependence of viscosity is used to adjust process variables for glass-making, from melting to annealing. The aim of this work was to develop a physics-informed machine learning model capable of predicting $η(T)$ of oxide liquids. Instead of predicting the viscosity itself, the NN predicts the parameters of the MYEGA viscosity equation: the liquid's fragility index, the glass transition temperature, and the asymptotic viscosity. With these parameters, $η$ can be computed at any temperature of interest, with the advantage of good extrapolation capabilities inherent to the MYEGA equation. The dataset was collected from the SciGlass database; only oxide liquids with enough data points in the high and low viscosity regions were selected, resulting in a final dataset with 17,584 data points containing 847 different liquids. About 600 features were engineered from the liquids' chemical composition and 35 of these features were selected using a feature selection protocol. The hyperparameter (HP) tuning of the NN was performed in a set of experiments using both random search and Bayesian strategies, where a total of 700 HP sets were tested. The most successful HP sets were further tested using 10-fold cross-validation, and the one with the lowest average validation loss was selected as the best set. The final trained NN was tested with a test dataset of 85 liquids with different compositions than those used for training and validating the NN. The $R^2$ for the test dataset's prediction was 0.97. This work introduces three advantages: the model can predict viscosity as well as the liquids' glass transition temperature and fragility index; the model is designed and trained with a focus on extrapolation; finally, the model is available as free and open-source software licensed under the GPL3.
△ Less
Submitted 15 March, 2021; v1 submitted 7 July, 2020;
originally announced July 2020.
-
Which glass stability parameters can assess the glass-forming ability of oxide systems?
Authors:
Jeanini Jiusti,
Daniel R. Cassar,
Edgar D. Zanotto
Abstract:
Glass forming ability (GFA) is a property of utmost importance in glass science and technology. In this paper, we used a statistical methodology---involving bootstrap sampling and the Wilcoxon test---to find out which glass stability parameters can better predict the glass forming ability. We collected or measured the necessary data for twelve stoichiometric oxide glasses that underwent predominan…
▽ More
Glass forming ability (GFA) is a property of utmost importance in glass science and technology. In this paper, we used a statistical methodology---involving bootstrap sampling and the Wilcoxon test---to find out which glass stability parameters can better predict the glass forming ability. We collected or measured the necessary data for twelve stoichiometric oxide glasses that underwent predominant heterogeneous nucleation (the most common case). We found that some GS parameters could predict the GFA of these oxide glasses quite well, whereas others perform poorly. Parameter $K_w$ was the top ranked, closely followed by the $K_H$, $γ$, $H^{'}$, $ΔT_{rg}$, and $K_{cr}$. Our results corroborate previous reports carried out using a smaller number of glasses, much less GS parameters, and less rigorous statistics. We also found that using $T_c$ instead of $T_x$ improved the predictive power of these parameters. Finally, the Jezica, the only parameter considered here that predicts the GFA without requiring the production of a glass piece (i.e., without relying on any crystallization information), ranked reasonably well in our analysis.
△ Less
Submitted 4 January, 2020;
originally announced January 2020.
-
Critical assessment of the alleged failure of the Classical Nucleation Theory at low temperatures
Authors:
Daniel Roberto Cassar,
André Hofmeister Serra,
Oscar Peitl,
Edgar Dutra Zanotto
Abstract:
The Classical Nucleation Theory allegedly fails to describe the temperature dependence of the homogeneous crystal nucleation rates below the temperature of maximum nucleation, $T_{\mathrm{max}}$. Possible explanations for this suspected breakdown have been advanced in the literature. However, the simplest hypothesis has never been tested, that it is a byproduct of nucleation datasets that have not…
▽ More
The Classical Nucleation Theory allegedly fails to describe the temperature dependence of the homogeneous crystal nucleation rates below the temperature of maximum nucleation, $T_{\mathrm{max}}$. Possible explanations for this suspected breakdown have been advanced in the literature. However, the simplest hypothesis has never been tested, that it is a byproduct of nucleation datasets that have not reached the steady-state regime. In this work, we tested this possibility by analyzing published nucleation data for oxide supercooled liquids, using only nucleation and viscosity data measured in samples of the same glass batch that also have satisfied a steady-state regime test. Furthermore, all the uncertainty and regression confidence bands were computed and considered. Having this rigorous protocol, among the 6 datasets analyzed, we only found weak evidence supporting the existence of the nucleation break in 2 datasets. Our collective results thus indicate that the break at $T_{\mathrm{max}}$ is not a common feature of all glass-formers.
△ Less
Submitted 20 July, 2020; v1 submitted 8 February, 2019;
originally announced February 2019.
-
Solving the Classical Nucleation Theory with respect to the surface energy
Authors:
Daniel R. Cassar
Abstract:
An essential parameter of the Classical Nucleation Theory (CNT) is the surface energy between a critical-size nucleus and the ambient phase, $σ$. In condensed matter, this parameter cannot be experimentally determined independently of CNT. A common practice to obtain $σ$ is to assume a model for its temperature-dependence and perform a regression of the CNT equation against experimental nucleation…
▽ More
An essential parameter of the Classical Nucleation Theory (CNT) is the surface energy between a critical-size nucleus and the ambient phase, $σ$. In condensed matter, this parameter cannot be experimentally determined independently of CNT. A common practice to obtain $σ$ is to assume a model for its temperature-dependence and perform a regression of the CNT equation against experimental nucleation data. The drawback of this practice is that assuming the temperature-dependence of $σ$ adds a bias to the analysis. Nonetheless, this practice is common because an analytical solution of the Classical Nucleation Theory with respect to $σ$ is not possible considering common expressions of this theory. In this article, a general numerical solution to this problem using the Lambert $W$ function is proposed, tested, and compared with typical regression methods. The major advantage of the proposed method is that there is no need to assume a model for the temperature-dependence of $σ$.
△ Less
Submitted 11 March, 2019; v1 submitted 15 December, 2018;
originally announced December 2018.
