Docente
Ilum-CNPEM
Daniel R. Cassar
Publicações
2026
Serbena, Daniel Rodrigo; Cieslack, Isabela Luiza Fraron; Ratis, Renan Cassiano; Chaves, Débora Van Putten; de Oliveira Filho, Sergio Servilha; Neves, Henrique; dos Santos, Fernando Sluchensci; Cassar, Daniel R.; da Silva, Weber Claudio; Bonini, Juliana Sartori
Artificial Intelligence for Diagnosis, Risk Stratification, and Prognosis of Neuroblastoma - A Systematic Review and Meta-Analysis Journal Article
Em: Curr Oncol Rep, vol. 28, não 1, 2026, ISSN: 1534-6269.
@article{Serbena2026,
title = {Artificial Intelligence for Diagnosis, Risk Stratification, and Prognosis of Neuroblastoma - A Systematic Review and Meta-Analysis},
author = {Daniel Rodrigo Serbena and Isabela Luiza Fraron Cieslack and Renan Cassiano Ratis and Débora Van Putten Chaves and Sergio Servilha de Oliveira Filho and Henrique Neves and Fernando Sluchensci dos Santos and Daniel R. Cassar and Weber Claudio da Silva and Juliana Sartori Bonini},
doi = {10.1007/s11912-026-01824-0},
issn = {1534-6269},
year = {2026},
date = {2026-09-04},
urldate = {2026-12-00},
journal = {Curr Oncol Rep},
volume = {28},
number = {1},
publisher = {Springer Science and Business Media LLC},
abstract = {<jats:title>Abstract</jats:title>
<jats:sec>
<jats:title>Purpose</jats:title>
<jats:p>To synthesizes evidence on artificial intelligence (AI) performance in neuroblastoma (NB) diagnosis, risk stratification, prognosis, and genomic characterization.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Materials and methods</jats:title>
<jats:p>A systematic review and meta-analysis was conducted following PRISMA 2020 guidelines (PROSPERO: CRD42024539475) across five databases. Meta-analyses used random-effects models with logit-transformed Area Under the Curve (AUCs) and cluster-robust standard errors. AI models were classified as Machine Learning Models (MLM) or Hybrid Nomograms (HN) based on their construction methodology.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Results</jats:title>
<jats:p>Of 3,742 articles identified, 53 were included. MLMs demonstrated higher point estimates than radiologists in differential diagnosis (AUC: 0.87 vs. 0.83), though this difference was not statistically significant and carried substantial uncertainty. HNs achieved stronger performance in risk stratification (AUC: 0.87). AI-derived nomograms (AUC: 0.9) and gene signatures (AUC: 0.8) outperformed conventional prognostic markers descriptively. Chemotherapy response prediction remained below clinical utility thresholds across all model types. Only 33.9% of models reported calibration and 24.5% underwent external validation.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Conclusions</jats:title>
<jats:p>AI demonstrates proof-of-concept across multiple NB clinical domains. However, clinical adoption remains premature given persistent gaps in external validation, calibration, dataset size, and pediatric-specific model development. Future studies should test these models prospectively in multicenter pediatric cohorts, ideally through COG or SIOPEN, using shared definitions for diagnosis, risk group, treatment response, and survival outcomes.</jats:p>
</jats:sec>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
<jats:title>Abstract</jats:title>
<jats:sec>
<jats:title>Purpose</jats:title>
<jats:p>To synthesizes evidence on artificial intelligence (AI) performance in neuroblastoma (NB) diagnosis, risk stratification, prognosis, and genomic characterization.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Materials and methods</jats:title>
<jats:p>A systematic review and meta-analysis was conducted following PRISMA 2020 guidelines (PROSPERO: CRD42024539475) across five databases. Meta-analyses used random-effects models with logit-transformed Area Under the Curve (AUCs) and cluster-robust standard errors. AI models were classified as Machine Learning Models (MLM) or Hybrid Nomograms (HN) based on their construction methodology.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Results</jats:title>
<jats:p>Of 3,742 articles identified, 53 were included. MLMs demonstrated higher point estimates than radiologists in differential diagnosis (AUC: 0.87 vs. 0.83), though this difference was not statistically significant and carried substantial uncertainty. HNs achieved stronger performance in risk stratification (AUC: 0.87). AI-derived nomograms (AUC: 0.9) and gene signatures (AUC: 0.8) outperformed conventional prognostic markers descriptively. Chemotherapy response prediction remained below clinical utility thresholds across all model types. Only 33.9% of models reported calibration and 24.5% underwent external validation.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Conclusions</jats:title>
<jats:p>AI demonstrates proof-of-concept across multiple NB clinical domains. However, clinical adoption remains premature given persistent gaps in external validation, calibration, dataset size, and pediatric-specific model development. Future studies should test these models prospectively in multicenter pediatric cohorts, ideally through COG or SIOPEN, using shared definitions for diagnosis, risk group, treatment response, and survival outcomes.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Purpose</jats:title>
<jats:p>To synthesizes evidence on artificial intelligence (AI) performance in neuroblastoma (NB) diagnosis, risk stratification, prognosis, and genomic characterization.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Materials and methods</jats:title>
<jats:p>A systematic review and meta-analysis was conducted following PRISMA 2020 guidelines (PROSPERO: CRD42024539475) across five databases. Meta-analyses used random-effects models with logit-transformed Area Under the Curve (AUCs) and cluster-robust standard errors. AI models were classified as Machine Learning Models (MLM) or Hybrid Nomograms (HN) based on their construction methodology.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Results</jats:title>
<jats:p>Of 3,742 articles identified, 53 were included. MLMs demonstrated higher point estimates than radiologists in differential diagnosis (AUC: 0.87 vs. 0.83), though this difference was not statistically significant and carried substantial uncertainty. HNs achieved stronger performance in risk stratification (AUC: 0.87). AI-derived nomograms (AUC: 0.9) and gene signatures (AUC: 0.8) outperformed conventional prognostic markers descriptively. Chemotherapy response prediction remained below clinical utility thresholds across all model types. Only 33.9% of models reported calibration and 24.5% underwent external validation.</jats:p>
</jats:sec>
<jats:sec>
<jats:title>Conclusions</jats:title>
<jats:p>AI demonstrates proof-of-concept across multiple NB clinical domains. However, clinical adoption remains premature given persistent gaps in external validation, calibration, dataset size, and pediatric-specific model development. Future studies should test these models prospectively in multicenter pediatric cohorts, ideally through COG or SIOPEN, using shared definitions for diagnosis, risk group, treatment response, and survival outcomes.</jats:p>
</jats:sec>
Vitoria, Leonardo Santos; Nascimento, Marcio Luis Ferreira; Lalic, Susana Souza; Cassar, Daniel R.
Physics-Informed Symbolic Regression for Predicting the Glass Transition Temperature of Alkali Borate Glasses Miscellaneous
2026.
@misc{vitoria2026physicsinformedsymbolicregressionpredicting,
title = {Physics-Informed Symbolic Regression for Predicting the Glass Transition Temperature of Alkali Borate Glasses},
author = {Leonardo Santos Vitoria and Marcio Luis Ferreira Nascimento and Susana Souza Lalic and Daniel R. Cassar},
url = {https://arxiv.org/abs/2608.14853},
year = {2026},
date = {2026-08-14},
urldate = {2026-01-01},
abstract = {The glass transition temperature (Tg) of alkali borate glasses is strongly composition-dependent and difficult to predict from first principles due to the structural complexity of the boron network. Here, we apply physics-informed symbolic regression (combining evolutive search with physically meaningful descriptors) to derive an interpretable closed-form expression for Tg in the xM2O⋅(100−x)B2O3 glass family, with M = Li, Na, and K and x expressed in mol%, and subsequently extrapolate it to M = Rb and Cs. The resulting model achieves a root-mean-square error of 14-16 K while maintaining clear physical interpretability, explicitly capturing the interplay among Tg, structural dissociation energy, and network packing. Critically, models built on the Rigid Unit Packing Fraction (RUPF) yield substantially more realistic Tg predictions than those using the conventional Atomic Packing Fraction (APF), as APF overestimates structural rigidity at intermediate compositions. The fitted dissociation energies are further validated against the revised Makishima-Mackenzie model, confirming that the inferred parameters are physically consistent, not merely statistically effective, within the alkali borate family. Finally, Monte Carlo uncertainty quantification reveals that prediction uncertainty is highest in the compositional regions associated with the boron anomaly, directly linking model limitations to a known structural transition in these glasses. This result highlights the potential of physics-informed symbolic regression as a transparent and interpretable alternative to black-box models for property prediction in glass systems.},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
The glass transition temperature (Tg) of alkali borate glasses is strongly composition-dependent and difficult to predict from first principles due to the structural complexity of the boron network. Here, we apply physics-informed symbolic regression (combining evolutive search with physically meaningful descriptors) to derive an interpretable closed-form expression for Tg in the xM2O⋅(100−x)B2O3 glass family, with M = Li, Na, and K and x expressed in mol%, and subsequently extrapolate it to M = Rb and Cs. The resulting model achieves a root-mean-square error of 14-16 K while maintaining clear physical interpretability, explicitly capturing the interplay among Tg, structural dissociation energy, and network packing. Critically, models built on the Rigid Unit Packing Fraction (RUPF) yield substantially more realistic Tg predictions than those using the conventional Atomic Packing Fraction (APF), as APF overestimates structural rigidity at intermediate compositions. The fitted dissociation energies are further validated against the revised Makishima-Mackenzie model, confirming that the inferred parameters are physically consistent, not merely statistically effective, within the alkali borate family. Finally, Monte Carlo uncertainty quantification reveals that prediction uncertainty is highest in the compositional regions associated with the boron anomaly, directly linking model limitations to a known structural transition in these glasses. This result highlights the potential of physics-informed symbolic regression as a transparent and interpretable alternative to black-box models for property prediction in glass systems.
Vitoria, Leonardo Santos; Nascimento, Marcio Luis Ferreira; Lalic, Susana Souza; Cassar, Daniel R.
Physics-Informed Symbolic Regression for Predicting the Glass Transition Temperature of Alkali Borate Glasses Miscellaneous
2026.
@misc{vitoria2026physicsinformedsymbolicregressionpredictingb,
title = {Physics-Informed Symbolic Regression for Predicting the Glass Transition Temperature of Alkali Borate Glasses},
author = {Leonardo Santos Vitoria and Marcio Luis Ferreira Nascimento and Susana Souza Lalic and Daniel R. Cassar},
url = {https://arxiv.org/abs/2608.14853},
year = {2026},
date = {2026-08-14},
urldate = {2026-01-01},
abstract = {The glass transition temperature (Tg) of alkali borate glasses is strongly composition-dependent and difficult to predict from first principles due to the structural complexity of the boron network. Here, we apply physics-informed symbolic regression (combining evolutive search with physically meaningful descriptors) to derive an interpretable closed-form expression for Tg in the xM2O⋅(100−x)B2O3 glass family, with M = Li, Na, and K and x expressed in mol%, and subsequently extrapolate it to M = Rb and Cs. The resulting model achieves a root-mean-square error of 14-16 K while maintaining clear physical interpretability, explicitly capturing the interplay among Tg, structural dissociation energy, and network packing. Critically, models built on the Rigid Unit Packing Fraction (RUPF) yield substantially more realistic Tg predictions than those using the conventional Atomic Packing Fraction (APF), as APF overestimates structural rigidity at intermediate compositions. The fitted dissociation energies are further validated against the revised Makishima-Mackenzie model, confirming that the inferred parameters are physically consistent, not merely statistically effective, within the alkali borate family. Finally, Monte Carlo uncertainty quantification reveals that prediction uncertainty is highest in the compositional regions associated with the boron anomaly, directly linking model limitations to a known structural transition in these glasses. This result highlights the potential of physics-informed symbolic regression as a transparent and interpretable alternative to black-box models for property prediction in glass systems.},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
The glass transition temperature (Tg) of alkali borate glasses is strongly composition-dependent and difficult to predict from first principles due to the structural complexity of the boron network. Here, we apply physics-informed symbolic regression (combining evolutive search with physically meaningful descriptors) to derive an interpretable closed-form expression for Tg in the xM2O⋅(100−x)B2O3 glass family, with M = Li, Na, and K and x expressed in mol%, and subsequently extrapolate it to M = Rb and Cs. The resulting model achieves a root-mean-square error of 14-16 K while maintaining clear physical interpretability, explicitly capturing the interplay among Tg, structural dissociation energy, and network packing. Critically, models built on the Rigid Unit Packing Fraction (RUPF) yield substantially more realistic Tg predictions than those using the conventional Atomic Packing Fraction (APF), as APF overestimates structural rigidity at intermediate compositions. The fitted dissociation energies are further validated against the revised Makishima-Mackenzie model, confirming that the inferred parameters are physically consistent, not merely statistically effective, within the alkali borate family. Finally, Monte Carlo uncertainty quantification reveals that prediction uncertainty is highest in the compositional regions associated with the boron anomaly, directly linking model limitations to a known structural transition in these glasses. This result highlights the potential of physics-informed symbolic regression as a transparent and interpretable alternative to black-box models for property prediction in glass systems.
Souza, Maurício Lima; Santos, Gisele Guimarães; Cassar, Daniel Roberto; Zanotto, Edgar Dutra
Machine-learning assisted design of lead-free niobium crystal glass with improved optical properties Journal Article
Em: Ceramics International, vol. 52, não 11, pp. 16129–16142, 2026, ISSN: 0272-8842.
@article{LimaSouza2026,
title = {Machine-learning assisted design of lead-free niobium crystal glass with improved optical properties},
author = {Maurício Lima Souza and Gisele Guimarães Santos and Daniel Roberto Cassar and Edgar Dutra Zanotto},
doi = {10.1016/j.ceramint.2026.02.212},
issn = {0272-8842},
year = {2026},
date = {2026-05-01},
journal = {Ceramics International},
volume = {52},
number = {11},
pages = {16129–16142},
publisher = {Elsevier BV},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Carvalho, Diogo P. L.; Loponi, Ana C. B.; Cassar, Daniel R.
Will it form a glass? Tackling glass formation using binary classification Miscellaneous
2026.
@misc{carvalho2026formglasstacklingglass,
title = {Will it form a glass? Tackling glass formation using binary classification},
author = {Diogo P. L. Carvalho and Ana C. B. Loponi and Daniel R. Cassar},
url = {https://arxiv.org/abs/2603.15312},
year = {2026},
date = {2026-03-16},
urldate = {2026-01-01},
abstract = {Glass formation is one of the most important and fundamental open problems in glass science. Predicting whether a liquid can be easily frozen into a glass appears simple but is far from it. In this communication, we address glass formation in inorganic nonmetallic liquids using binary classification to predict the probability that a given liquid will form a glass under typical laboratory conditions. Using a dataset of more than 50,000 examples, we trained random forest classifiers that achieved ROC-AUC values around 0.89 and PR-AUC close to 0.95 on the holdout dataset (i.e., unseen data). A rigorous model selection routine was employed, including hyperparameter tuning with cross-validation, and four different data treatment routes were evaluated. Using SHAP values, we extracted valuable insights from the trained models that both agree with established knowledge and extend it. For example, we identified that the bandgap energy of the constituent chemical elements is positively correlated with glass formation. When glass stability parameters and Jezica were added to the dataset, no performance improvement was observed, but model complexity decreased significantly. This result is particularly relevant for composition screening, especially in inverse design problems.},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Glass formation is one of the most important and fundamental open problems in glass science. Predicting whether a liquid can be easily frozen into a glass appears simple but is far from it. In this communication, we address glass formation in inorganic nonmetallic liquids using binary classification to predict the probability that a given liquid will form a glass under typical laboratory conditions. Using a dataset of more than 50,000 examples, we trained random forest classifiers that achieved ROC-AUC values around 0.89 and PR-AUC close to 0.95 on the holdout dataset (i.e., unseen data). A rigorous model selection routine was employed, including hyperparameter tuning with cross-validation, and four different data treatment routes were evaluated. Using SHAP values, we extracted valuable insights from the trained models that both agree with established knowledge and extend it. For example, we identified that the bandgap energy of the constituent chemical elements is positively correlated with glass formation. When glass stability parameters and Jezica were added to the dataset, no performance improvement was observed, but model complexity decreased significantly. This result is particularly relevant for composition screening, especially in inverse design problems.
2025
Silva, Roni A.; Batista, Gislene; Bradtmüller, Henrik; Campos, João V.; Martins, Gabriela K.; Cassar, Daniel R.; Kurelo, Bruna C. E. S.; Zallocco, Vinicius M.; Rodrigues, Ana C. M.; Cassanjes, Fabia C.; Poirier, Gael Y.; Serbena, Francisco C.
Structure–property relationships in sodium phosphate glasses and glass‐ceramics containing tantalum oxide Journal Article
Em: J Am Ceram Soc., 2025, ISSN: 1551-2916.
@article{Silva2025,
title = {Structure–property relationships in sodium phosphate glasses and glass‐ceramics containing tantalum oxide},
author = {Roni A. Silva and Gislene Batista and Henrik Bradtmüller and João V. Campos and Gabriela K. Martins and Daniel R. Cassar and Bruna C. E. S. Kurelo and Vinicius M. Zallocco and Ana C. M. Rodrigues and Fabia C. Cassanjes and Gael Y. Poirier and Francisco C. Serbena},
doi = {10.1111/jace.20677},
issn = {1551-2916},
year = {2025},
date = {2025-05-20},
urldate = {2025-05-20},
journal = {J Am Ceram Soc.},
publisher = {Wiley},
abstract = {<jats:title>Abstract</jats:title><jats:p>In this study, we examine the impact of structural modifications from incorporating tantalum oxide on the thermal, electrical, and mechanical properties of sodium phosphate glasses and glass‐ceramics. Although these materials are well known for optical applications, this work aims to systematically explore their macroscopic properties, which are yet to be fully characterized. Glass samples were prepared in the binary molar system (100 − <jats:italic>x</jats:italic>)NaPO<jats:sub>3</jats:sub>–<jats:italic>x</jats:italic>Ta<jats:sub>2</jats:sub>O<jats:sub>5</jats:sub> with <jats:italic>x </jats:italic>= 20, 30, 40, 47.5, 50. A transparent glass‐ceramic was also produced by heat‐treating the 52.5NaPO<jats:sub>3</jats:sub>–47.5Ta<jats:sub>2</jats:sub>O<jats:sub>5</jats:sub> glass composition. Structural characterization was performed by Raman, Fourier transform infrared (FTIR), and solid‐state nuclear magnetic resonance (NMR) spectroscopies. Increasing tantalum oxide content led to a notable increase in glass transition temperature together with a reduced thermal stability against crystallization, indicating higher glass network connectivity at higher tantalum levels. Thermal analysis and X‐ray diffraction (XRD) confirmed the formation of a single crystalline phase in the glass‐ceramic, identified as the bronze‐like perovskite Na<jats:sub>2</jats:sub>Ta<jats:sub>8</jats:sub>O<jats:sub>21</jats:sub> with an average particle size of 31 nm. Electrical properties were investigated using impedance and electric modulus formalisms, revealing that higher tantalum content increases resistivity and decreases conductivity, attributed to reduced Na<jats:sup>+</jats:sup> ion concentration and increased atomic packing density. Interestingly, glass‐ceramics exhibited slightly higher conductivity than pristine glass. Density, Vickers hardness, Young's modulus, and nanoindentation hardness also increased significantly with higher tantalum content, while crystallization had a minimal effect on these properties. Overall, these results indicate that higher tantalum oxide content not only enhances the glass network's connectivity but also significantly influences the electrical and mechanical properties of sodium phosphate glasses and glass‐ceramics.</jats:p>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
<jats:title>Abstract</jats:title><jats:p>In this study, we examine the impact of structural modifications from incorporating tantalum oxide on the thermal, electrical, and mechanical properties of sodium phosphate glasses and glass‐ceramics. Although these materials are well known for optical applications, this work aims to systematically explore their macroscopic properties, which are yet to be fully characterized. Glass samples were prepared in the binary molar system (100 − <jats:italic>x</jats:italic>)NaPO<jats:sub>3</jats:sub>–<jats:italic>x</jats:italic>Ta<jats:sub>2</jats:sub>O<jats:sub>5</jats:sub> with <jats:italic>x </jats:italic>= 20, 30, 40, 47.5, 50. A transparent glass‐ceramic was also produced by heat‐treating the 52.5NaPO<jats:sub>3</jats:sub>–47.5Ta<jats:sub>2</jats:sub>O<jats:sub>5</jats:sub> glass composition. Structural characterization was performed by Raman, Fourier transform infrared (FTIR), and solid‐state nuclear magnetic resonance (NMR) spectroscopies. Increasing tantalum oxide content led to a notable increase in glass transition temperature together with a reduced thermal stability against crystallization, indicating higher glass network connectivity at higher tantalum levels. Thermal analysis and X‐ray diffraction (XRD) confirmed the formation of a single crystalline phase in the glass‐ceramic, identified as the bronze‐like perovskite Na<jats:sub>2</jats:sub>Ta<jats:sub>8</jats:sub>O<jats:sub>21</jats:sub> with an average particle size of 31 nm. Electrical properties were investigated using impedance and electric modulus formalisms, revealing that higher tantalum content increases resistivity and decreases conductivity, attributed to reduced Na<jats:sup>+</jats:sup> ion concentration and increased atomic packing density. Interestingly, glass‐ceramics exhibited slightly higher conductivity than pristine glass. Density, Vickers hardness, Young's modulus, and nanoindentation hardness also increased significantly with higher tantalum content, while crystallization had a minimal effect on these properties. Overall, these results indicate that higher tantalum oxide content not only enhances the glass network's connectivity but also significantly influences the electrical and mechanical properties of sodium phosphate glasses and glass‐ceramics.</jats:p>
2024
Stoco, Caroline Binde; Cassar, Daniel R.; Santana, Geovana Lira; Kaufman, Michael; Clarke, Amy; Coury, Francisco Gil
Optimizing toughness in high entropy alloys using a genetic algorithm: A combined computational and experimental approach Journal Article
Em: Materials Today Communications, vol. 41, 2024, ISSN: 2352-4928.
@article{Stoco2024,
title = {Optimizing toughness in high entropy alloys using a genetic algorithm: A combined computational and experimental approach},
author = {Caroline Binde Stoco and Daniel R. Cassar and Geovana Lira Santana and Michael Kaufman and Amy Clarke and Francisco Gil Coury},
doi = {10.1016/j.mtcomm.2024.110768},
issn = {2352-4928},
year = {2024},
date = {2024-12-01},
journal = {Materials Today Communications},
volume = {41},
publisher = {Elsevier BV},
abstract = {A genetic algorithm (GA) was developed to search for high entropy alloys (HEAs) with good combinations of mechanical properties. The algorithm was designed find single-phase face-centered cubic (FCC) HEAs, already known for their ductility, with high Hall-Petch constants (K) and high critical resolved shear stresses (
). The objective was to develop a methodology that allows the design of HEAs in a multi-objective environment, focusing on alloys that exhibit enhanced ductility and strength, achieved through an increase in its yield point without significant loss of its ultimate deformation via adjustments of K and
values, resulting in an alloy with high toughness. The most promising alloy suggested by the genetic algorithm was an unconventional composition (Co32.73Cu15.11Fe0.72Hf0.72Mn35.97Mo3.96Ni10.43Sn0.36), with eight different elements in non-equiatomic ratios with maximized K and
values. This alloy was then experimentally produced and characterized by scanning electron microscopy (SEM), synchrotron x-ray diffraction (XRD), transmission electron microscopy (TEM) and microhardness, with the objective of validating the predictions made by the GA. An advantage of the proposed method is the possibility of more systematically identifying and exploring new compositions in the complex composition space characteristic of these multicomponent alloys, as it directs its search towards domains that can be used to solve challenging problems involving multi-objective optimization. However, the thermodynamic parameters used for single-phase FCC prediction, namely the valence electron concentration (VEC) and the dimensionless thermodynamic parameter φ, exhibited limitations. These limitations were further explored in the present study, as evidenced by the fact that the microstructure of the selected alloy was not single-phase, which hindered the study of the mechanical properties, predicted exclusively for monophasic FCC alloys. Therefore, this work highlights the necessity for the development of novel thermodynamic equations for phase prediction, or even the integration of the genetic algorithm with other methodologies, such as CALPHAD calculations, to achieve enhanced results.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
A genetic algorithm (GA) was developed to search for high entropy alloys (HEAs) with good combinations of mechanical properties. The algorithm was designed find single-phase face-centered cubic (FCC) HEAs, already known for their ductility, with high Hall-Petch constants (K) and high critical resolved shear stresses (
). The objective was to develop a methodology that allows the design of HEAs in a multi-objective environment, focusing on alloys that exhibit enhanced ductility and strength, achieved through an increase in its yield point without significant loss of its ultimate deformation via adjustments of K and
values, resulting in an alloy with high toughness. The most promising alloy suggested by the genetic algorithm was an unconventional composition (Co32.73Cu15.11Fe0.72Hf0.72Mn35.97Mo3.96Ni10.43Sn0.36), with eight different elements in non-equiatomic ratios with maximized K and
values. This alloy was then experimentally produced and characterized by scanning electron microscopy (SEM), synchrotron x-ray diffraction (XRD), transmission electron microscopy (TEM) and microhardness, with the objective of validating the predictions made by the GA. An advantage of the proposed method is the possibility of more systematically identifying and exploring new compositions in the complex composition space characteristic of these multicomponent alloys, as it directs its search towards domains that can be used to solve challenging problems involving multi-objective optimization. However, the thermodynamic parameters used for single-phase FCC prediction, namely the valence electron concentration (VEC) and the dimensionless thermodynamic parameter φ, exhibited limitations. These limitations were further explored in the present study, as evidenced by the fact that the microstructure of the selected alloy was not single-phase, which hindered the study of the mechanical properties, predicted exclusively for monophasic FCC alloys. Therefore, this work highlights the necessity for the development of novel thermodynamic equations for phase prediction, or even the integration of the genetic algorithm with other methodologies, such as CALPHAD calculations, to achieve enhanced results.
). The objective was to develop a methodology that allows the design of HEAs in a multi-objective environment, focusing on alloys that exhibit enhanced ductility and strength, achieved through an increase in its yield point without significant loss of its ultimate deformation via adjustments of K and
values, resulting in an alloy with high toughness. The most promising alloy suggested by the genetic algorithm was an unconventional composition (Co32.73Cu15.11Fe0.72Hf0.72Mn35.97Mo3.96Ni10.43Sn0.36), with eight different elements in non-equiatomic ratios with maximized K and
values. This alloy was then experimentally produced and characterized by scanning electron microscopy (SEM), synchrotron x-ray diffraction (XRD), transmission electron microscopy (TEM) and microhardness, with the objective of validating the predictions made by the GA. An advantage of the proposed method is the possibility of more systematically identifying and exploring new compositions in the complex composition space characteristic of these multicomponent alloys, as it directs its search towards domains that can be used to solve challenging problems involving multi-objective optimization. However, the thermodynamic parameters used for single-phase FCC prediction, namely the valence electron concentration (VEC) and the dimensionless thermodynamic parameter φ, exhibited limitations. These limitations were further explored in the present study, as evidenced by the fact that the microstructure of the selected alloy was not single-phase, which hindered the study of the mechanical properties, predicted exclusively for monophasic FCC alloys. Therefore, this work highlights the necessity for the development of novel thermodynamic equations for phase prediction, or even the integration of the genetic algorithm with other methodologies, such as CALPHAD calculations, to achieve enhanced results.
Acosta, María Helena Ramírez; Cassar, Daniel R.; Rodrigues, Lorena Raphael; Baldin, João Marcos Conradi; Zanotto, Edgar Dutra
Diffusion proxies reveal the dynamic process in supercooled and glassy lithium diborate Journal Article
Em: Ceramics International, 2024, ISSN: 0272-8842.
@article{RAMIREZACOSTA2024,
title = {Diffusion proxies reveal the dynamic process in supercooled and glassy lithium diborate},
author = {María Helena Ramírez Acosta and Daniel R. Cassar and Lorena Raphael Rodrigues and João Marcos Conradi Baldin and Edgar Dutra Zanotto},
url = {https://www.sciencedirect.com/science/article/pii/S0272884224028153},
doi = {https://doi.org/10.1016/j.ceramint.2024.06.369},
issn = {0272-8842},
year = {2024},
date = {2024-06-27},
urldate = {2024-01-01},
journal = {Ceramics International},
abstract = {Understanding the effective diffusion coefficient (D) is crucial for describing atomic transport during crystal nucleation in supercooled liquids and glasses. However, pinpointing the key structural units driving crystallization in intricate, multicomponent glass-formers remains a challenge. This study presents a novel analysis of lithium diborate (Li₂O·2 B₂O₃ - LB₂) crystallization in supercooled and glassy states by using available viscosity and crystallization data for samples of the same batch. An original key feature is that the nucleation rates were measured using a single-stage rather than the traditional double-stage heat treatment. We compared three proxies for D: Dη obtained from viscosity, DU derived from crystal growth rates, and Dτ estimated from nucleation time-lags. Our analysis revealed that at deep supercoolings, below the glass transition temperature, DU and Dτ yield similar steady-state nucleation rate estimates, which significantly exceed those predicted by using Dη (the most frequently used diffusion proxy). This result suggests that the atomic jumps governing crystal nucleation may be comparable to those in crystal growth, but distinct from those associated with cooperative rearrangements controlling viscous flow. Additionally, the estimated kinetic spinodal temperature (Tks) is above the Kauzmann temperature Tk, implying that crystal nucleation precedes relaxation to the supercooled liquid state, circumventing the entropy paradox and corroborating recent MD simulations for other substances. These findings are valuable for refining theoretical models of crystal nucleation and highlight the complexities of the glassy state.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Understanding the effective diffusion coefficient (D) is crucial for describing atomic transport during crystal nucleation in supercooled liquids and glasses. However, pinpointing the key structural units driving crystallization in intricate, multicomponent glass-formers remains a challenge. This study presents a novel analysis of lithium diborate (Li₂O·2 B₂O₃ - LB₂) crystallization in supercooled and glassy states by using available viscosity and crystallization data for samples of the same batch. An original key feature is that the nucleation rates were measured using a single-stage rather than the traditional double-stage heat treatment. We compared three proxies for D: Dη obtained from viscosity, DU derived from crystal growth rates, and Dτ estimated from nucleation time-lags. Our analysis revealed that at deep supercoolings, below the glass transition temperature, DU and Dτ yield similar steady-state nucleation rate estimates, which significantly exceed those predicted by using Dη (the most frequently used diffusion proxy). This result suggests that the atomic jumps governing crystal nucleation may be comparable to those in crystal growth, but distinct from those associated with cooperative rearrangements controlling viscous flow. Additionally, the estimated kinetic spinodal temperature (Tks) is above the Kauzmann temperature Tk, implying that crystal nucleation precedes relaxation to the supercooled liquid state, circumventing the entropy paradox and corroborating recent MD simulations for other substances. These findings are valuable for refining theoretical models of crystal nucleation and highlight the complexities of the glassy state.
Allec, Sarah I.; Lu, Xiaonan; Cassar, Daniel R.; Nguyen, Xuan T.; Hegde, Vinay I.; Mahadevan, Thiruvillamalai; Peterson, Miroslava; Du, Jincheng; Riley, Brian J.; Vienna, John D.; Saal, James E.
Evaluation of GlassNet for physics-informed machine learning of glass stability and glass-forming ability Working paper
2024.
@workingpaper{allec2024evaluation,
title = {Evaluation of GlassNet for physics-informed machine learning of glass stability and glass-forming ability},
author = {Sarah I. Allec and Xiaonan Lu and Daniel R. Cassar and Xuan T. Nguyen and Vinay I. Hegde and Thiruvillamalai Mahadevan and Miroslava Peterson and Jincheng Du and Brian J. Riley and John D. Vienna and James E. Saal},
url = {https://arxiv.org/abs/2403.10682},
doi = {10.48550/arXiv.2403.10682},
year = {2024},
date = {2024-03-19},
urldate = {2024-01-01},
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 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.},
keywords = {},
pubstate = {published},
tppubtype = {workingpaper}
}
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.
Silva, Roni Alisson; Batista, Gislene; Cassani, Rodrigo; Teófilo, Ana Flávia; Martins, Gabriela Kobelnik; Cassar, Daniel R.; Serbena, Francisco Carlos; Cassanjes, Fábia; Poirier, Gael
Thermal, chemical, and mechanical properties of niobium phosphate glasses and glass-ceramics Journal Article
Em: Ceramics International, 2024, ISSN: 0272-8842.
@article{SILVA2024,
title = {Thermal, chemical, and mechanical properties of niobium phosphate glasses and glass-ceramics},
author = {Roni Alisson Silva and Gislene Batista and Rodrigo Cassani and Ana Flávia Teófilo and Gabriela Kobelnik Martins and Daniel R. Cassar and Francisco Carlos Serbena and Fábia Cassanjes and Gael Poirier},
url = {https://www.sciencedirect.com/science/article/pii/S0272884224008733},
doi = {https://doi.org/10.1016/j.ceramint.2024.02.350},
issn = {0272-8842},
year = {2024},
date = {2024-02-29},
urldate = {2024-01-01},
journal = {Ceramics International},
abstract = {In this work, glass samples were obtained by melt-quenching in the binary system (100-x)KPO3-xNb2O5 with x = 20, 30, 40, and 50 mol%. Thermal properties investigated by DSC together with structural investigations by Raman spectroscopy allowed to understand the structural effect of Nb2O5 incorporation in the potassium phosphate glass network. A transparent glass-ceramic has also been produced by heat treatment of the 50KPO3–50Nb2O5 glass sample. Density, molar volume, atomic packing, and refractive index were determined and increased with Nb2O5 content. Chemical resistance in several corrosive aqueous solutions was found to increase with Nb2O5 content and crystallization. Vickers hardness as well as elastic modulus and hardness obtained from nanoindentation experiments also strongly increased with higher niobium contents, indicating that not only optical properties but also chemical and mechanical properties are improved with niobium incorporation in the phosphate glass network.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In this work, glass samples were obtained by melt-quenching in the binary system (100-x)KPO3-xNb2O5 with x = 20, 30, 40, and 50 mol%. Thermal properties investigated by DSC together with structural investigations by Raman spectroscopy allowed to understand the structural effect of Nb2O5 incorporation in the potassium phosphate glass network. A transparent glass-ceramic has also been produced by heat treatment of the 50KPO3–50Nb2O5 glass sample. Density, molar volume, atomic packing, and refractive index were determined and increased with Nb2O5 content. Chemical resistance in several corrosive aqueous solutions was found to increase with Nb2O5 content and crystallization. Vickers hardness as well as elastic modulus and hardness obtained from nanoindentation experiments also strongly increased with higher niobium contents, indicating that not only optical properties but also chemical and mechanical properties are improved with niobium incorporation in the phosphate glass network.
Vitoria, Leonardo Santos; Cassar, Daniel R.; Lalic, Susana Souza; Nascimento, Marcio Luis Ferreira
Using a simple radial basis function neural network to predict the glass transition temperature of alkali borate glasses Journal Article
Em: Journal of Non-Crystalline Solids, vol. 629, pp. 122870, 2024, ISSN: 0022-3093.
@article{VITORIA2024122870,
title = {Using a simple radial basis function neural network to predict the glass transition temperature of alkali borate glasses},
author = {Leonardo Santos Vitoria and Daniel R. Cassar and Susana Souza Lalic and Marcio Luis Ferreira Nascimento},
url = {https://www.sciencedirect.com/science/article/pii/S0022309324000516},
doi = {https://doi.org/10.1016/j.jnoncrysol.2024.122870},
issn = {0022-3093},
year = {2024},
date = {2024-02-17},
urldate = {2024-01-01},
journal = {Journal of Non-Crystalline Solids},
volume = {629},
pages = {122870},
abstract = {Artificial Neural Networks (ANN) are powerful machine learning algorithms. In the context of glass science, ANNs have been successfully used to induce composition-property regression models. These models are already influencing the way that we design new glasses, moving away from the traditional “cook and look” approach to computer-aided design. In this work, we induced an ANN model for the glass transition temperature Tg for glasses of the family xM2O⋅(100−x)B2O3, where M = Li, Na, K, Rb, and Cs, using simple radial basis function neural networks (RBF) with only two neurons. Statistically, the results of our methodologies are comparable to other complex models (empirical or physical) available in the literature with more neurons, such as the recent GlassNet and the topological model proposed by Mauro and co-authors (2009). This result supports the idea that a simpler specialized model can perform better than a complex general model for a restricted training domain.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Artificial Neural Networks (ANN) are powerful machine learning algorithms. In the context of glass science, ANNs have been successfully used to induce composition-property regression models. These models are already influencing the way that we design new glasses, moving away from the traditional “cook and look” approach to computer-aided design. In this work, we induced an ANN model for the glass transition temperature Tg for glasses of the family xM2O⋅(100−x)B2O3, where M = Li, Na, K, Rb, and Cs, using simple radial basis function neural networks (RBF) with only two neurons. Statistically, the results of our methodologies are comparable to other complex models (empirical or physical) available in the literature with more neurons, such as the recent GlassNet and the topological model proposed by Mauro and co-authors (2009). This result supports the idea that a simpler specialized model can perform better than a complex general model for a restricted training domain.
2023
Cassar, Daniel R.
GlassNet: A multitask deep neural network for predicting many glass properties Journal Article
Em: Ceramics International, 2023.
@article{Cassar_2023,
title = {GlassNet: A multitask deep neural network for predicting many glass properties},
author = {Daniel R. Cassar},
url = {https://doi.org/10.1016%2Fj.ceramint.2023.08.281},
doi = {10.1016/j.ceramint.2023.08.281},
year = {2023},
date = {2023-08-01},
urldate = {2023-08-01},
journal = {Ceramics International},
publisher = {Elsevier BV},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Welch, Rebecca S.; Zanotto, Edgar D.; Wilkinson, Collin J.; Cassar, Daniel R.; Montazerian, Maziar; Mauro, John C.
Cracking the Kauzmann paradox Journal Article
Em: Acta Materialia, vol. 254, pp. 118994, 2023, ISSN: 1359-6454.
@article{WELCH2023118994,
title = {Cracking the Kauzmann paradox},
author = {Rebecca S. Welch and Edgar D. Zanotto and Collin J. Wilkinson and Daniel R. Cassar and Maziar Montazerian and John C. Mauro},
url = {https://www.sciencedirect.com/science/article/pii/S1359645423003257},
doi = {https://doi.org/10.1016/j.actamat.2023.118994},
issn = {1359-6454},
year = {2023},
date = {2023-01-01},
journal = {Acta Materialia},
volume = {254},
pages = {118994},
abstract = {The Kauzmann paradox and associated Kauzmann temperature are two of the most widely debated topics in glass science over the past eighty years. Both conceptualized by Walter Kauzmann in 1948, the Kauzmann paradox occurs when some supercooled liquids apparently exhibit a negative excess entropy at temperatures above absolute zero. The Kauzmann temperature is the temperature at which the excess entropy vanishes. This review provides a retrospective on the origin of these hypotheses and their study through energy landscapes, crystallization behavior, and viscosity models. We also provide a critical analysis of each approach. After nearly eighty years of research, there is no conclusive evidence that supports the concepts proposed by Kauzmann. As such, it can be concluded that future work should be focused elsewhere.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The Kauzmann paradox and associated Kauzmann temperature are two of the most widely debated topics in glass science over the past eighty years. Both conceptualized by Walter Kauzmann in 1948, the Kauzmann paradox occurs when some supercooled liquids apparently exhibit a negative excess entropy at temperatures above absolute zero. The Kauzmann temperature is the temperature at which the excess entropy vanishes. This review provides a retrospective on the origin of these hypotheses and their study through energy landscapes, crystallization behavior, and viscosity models. We also provide a critical analysis of each approach. After nearly eighty years of research, there is no conclusive evidence that supports the concepts proposed by Kauzmann. As such, it can be concluded that future work should be focused elsewhere.
Mannan, Sajid; Zaki, Mohd; Bishnoi, Suresh; Cassar, Daniel R.; Jiusti, Jeanini; Faria, Julio Cesar Ferreira; Christensen, Johan F. S.; Gosvami, Nitya Nand; Smedskjaer, Morten M.; Zanotto, Edgar Dutra; Krishnan, N. M. Anoop
Glass hardness: Predicting composition and load effects via symbolic reasoning-informed machine learning Journal Article
Em: Acta Materialia, vol. 255, pp. 119046, 2023, ISSN: 1359-6454.
@article{MANNAN2023119046,
title = {Glass hardness: Predicting composition and load effects via symbolic reasoning-informed machine learning},
author = {Sajid Mannan and Mohd Zaki and Suresh Bishnoi and Daniel R. Cassar and Jeanini Jiusti and Julio Cesar Ferreira Faria and Johan F. S. Christensen and Nitya Nand Gosvami and Morten M. Smedskjaer and Edgar Dutra Zanotto and N. M. Anoop Krishnan},
url = {https://www.sciencedirect.com/science/article/pii/S1359645423003774},
doi = {https://doi.org/10.1016/j.actamat.2023.119046},
issn = {1359-6454},
year = {2023},
date = {2023-01-01},
journal = {Acta Materialia},
volume = {255},
pages = {119046},
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 3,000 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's equation) to describe the ISE with ML to develop a framework that incorporates the symbolic equation 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 the composition effect and ISE in glasses, enabling efficient and accelerated discovery of new glass compositions with targeted hardness.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
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 3,000 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's equation) to describe the ISE with ML to develop a framework that incorporates the symbolic equation 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 the composition effect and ISE in glasses, enabling efficient and accelerated discovery of new glass compositions with targeted hardness.