A predictable glass forming ability expression by statistical learning and evolutionary intelligence
Introduction
The bulk metallic glasses (BMGs) offer attractive properties to consider them as potential material for many advanced technological applications [1]. However, the limited section thickness, which can be produced as glass, without significant crystallization, limits such possibilities. Many compositional design methodologies have been proposed and used to design the composition of the bulk metallic glasses, with marginal success. For instance, the compositions explored and established for high glass forming ability generally bear costly elements like Pd and rare earths [1]. Therefore, it is essential to search for low cost BMGs with high glass forming ability (GFA) to utilize the technological advantages offered by these glassy materials. The experimental exploration for the compositions with high glass forming ability is an expensive and tedious method. Therefore, as an alternative, it is proposed to predict the glass forming ability of an alloy composition, before its actual physical preparation.
The glass forming ability indicator expressions may be instrumental to formulate a better compositional design strategy. Nevertheless, different glass forming ability indicator expressions have been reported in terms of characteristic temperatures(Tg, Tx and Tl,whereTg stands for glass transition temperature, Tx-onset temperature of crystallization and Tl is offset temperature of melting)of the metallic glass [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21]. These characteristic temperatures based GFA indicator expressions are useful to estimate the glass forming ability once the metallic glass is physically prepared. These GFA expressions lack the predictive capability from the generic data of the alloy, i.e. they cannot be used to estimate the glass forming ability of a composition before actual preparation of alloy [12]. Despite of the intensive efforts from the metallic glass workers, quantifying as well as predicting the glass forming ability still remains an open question [12].
Also few alternative approaches have been reported in the literature. The Inoue's empirical rule [3], deep eutectic based compositional design [22], [23], structure and topologybased approach [24], [25], [26], data training of artificial neural network [27], [28] are to name a few effort made to develop a strategy for compositional design.Also the thermodynamics based approach [29], [30], [31], [32], [33], cluster line based approach [34], [35], [36], molecular dynamics simulation based approach [37], [38] and a golden mean analysis based approach [39] has been attempted to design the BMG compositions.However, these attempts have led to limited success due to one or other constraints.
Nevertheless, over the past few years the data driven approachesare gaining reasonable attention of the metallic glass workers to derive more plausible solution of the problem [40], [41], [42], [43], [44], [45].Therefore, development of data driven model-capable to predict glass forming ability of a composition may providebetter opportunity for combinatorial analysis to audit a wide range of species of BMG composition.In this work, a meta-modeling approach using principle component analysis to statistically model the new features,i.e. principal components, subsequently used as model input, to model the GFA expression using evolutionary intelligence of genetic programming has been presented.The prediction results of the proposed meta-model of GFA expression has been discussed as case study in few ternary alloy systems.
Section snippets
The meta-modeling methodology
The primary interest of this work is to employ the statistical learning ability of principle component analysis and the evolutionary intelligence of genetic programming in the meta-modeling of BMG data knowledge for evolution of GFA expression. It has been attempted to explore the relation between GFA of the alloy with the generic attribute of the constituting elements and their quantitative presence in the alloy. Eventually an eleven dimensional descriptors space has been described by means of
Glass formation ability expression development
The proposed meta-modeling approach has been implemented for the analysis of the experimentally reported critical diameter, Dmax, for 594 synthesized BMGs alloys. In the initial stage, the model computes the scores of the principal components (PC1 and PC2) which describe the alloys in two dimension at the cost of marginal loss of knowledge present in the original eleven dimensional data space as the atomic fraction weighted sum of the eleven generic attribute of the composition data of the
Conclusion
The proposed method of meta-modeling successfully developed a glass forming ability expression in terms of principal components which are linear combination of the atomic fraction weighted sum of the generic attribute of the compositional elements of the alloys. The evolved GFA expression, PGp, has been shown a superior correlation of R = 0.5769 on 594 BMGs with corresponding critical diameter.The evolved glass forming ability parameter bears predictive capability unlike the characteristic
References (50)
- et al.
Reduced glass transition temperature and glass forming ability of bulk glass forming alloys
J. Non Cryst. Solids
(2000) Stabilization of metallic supercooled liquid and bulk amorphous alloys
Acta Mater.
(2000)- et al.
A new glass-forming ability criterion for bulk metallic glasses
Acta Mater.
(2002) - et al.
Heating and cooling rate dependence of the parameters representing the glass forming ability in bulk metallic glasses
J. Non Cryst. Solids
(2005) A new criterion for evaluating the glass-forming ability of bulk metallic glasses
Mater. Sci. Eng. A
(2006)- et al.
A new criterion for the glass-forming ability of liquids
J. Non Cryst. Solids
(2007) - et al.
Recent progress in quantifying glass-forming ability of bulk metallic glasses
Intermetallics
(2007) - et al.
A new criterion for evaluating the glass-forming ability of bulk glass forming alloys
J. Alloys Compd.
(2008) A new criterion for predicting the glass-forming ability of bulk metallic glasses
J. Alloys Compd.
(2009)On the new criterion to assess the glass-forming ability of metallic alloys
Mater. Sci. Eng. A
(2009)
A critical analysis of the glass-forming ability of alloys
J. Non Cryst. Solids
A new parameter to evaluate the glass-forming ability of bulk metallic glasses
Mater. Sci. Eng. A
New glass forming ability criterion derived from cooling consideration
Intermetallics
Identify the best glass forming ability criterion
Intermetallics
Sensitivity of the Hruby, Lu–Liu, Fan, Yuan, and Long glass stability parameters to the change of the ratios of characteristic temperatures Tx/Tg and Tm/Tg
Thermochim. Acta
Which glass stability criterion is the best?
Thermochim. Acta
A new approach to understanding and measuring glass formation in bulk amorphous materials
Intermetallics
Predictability of bulk metallic glass forming ability using the criteria based on characteristic temperatures of alloys
Phys. B Condens. Matter
Evolution of glass forming ability indicator by genetic programming
Comput. Mater. Sci.
The efficient cluster packing model – an atomic structural model for metallic glasses
Acta Mater.
Topological criterion for metallic glass formation
Mater. Sci. Eng. A
Specific criteria for selection of alloy compositions for bulk metallic glasses
Scr. Mater.
Artificial neural network modeling for undercooled liquid region of glass forming alloys
Comput. Mater. Sci.
Prediction of critical cooling rate for glass forming alloys by artificial neural network
Mater. Des.
Identification of compositions with highest glass forming ability in multicomponent systems by thermodynamic and topological approaches
Mater. Sci. Eng. A
Cited by (18)
Thermodynamic evaluation and experimental verification of the glass forming ability of Cu-Zr-based alloys
2021, Journal of Non-Crystalline SolidsDiscovery of novel quaternary bulk metallic glasses using a developed correlation-based neural network approach
2021, Computational Materials ScienceCitation Excerpt :For instance, Majid et al [27] reported that general regression neural network can accurately predict the GFA in the BMGs; however, their work lacked sufficient input data and ignored thermodynamic aspect of BMGs. Tripathi et al [28,29] developed a statistical learning model to express the GFA in terms of the basic components related to the generic attribute of the compositional elements. Using a vector classification method, Sun et al [30] found that the liquidus temperature fulfils a key role in the prediction of GFA for binary amorphous alloys.
Effect of network formers and modifiers on the crystallization resistance of oxide glasses
2020, Journal of Non-Crystalline SolidsCitation Excerpt :The determination of GFA is very time-consuming by conventional methods, such as the time-temperature-transformation (TTT) curve [17], the thermocouple method [18] or the Colmenero and Barandiarán method [19]. To avoid this problem, many simple parameters based on the characteristic temperatures of glasses and supercooled liquids have been suggested [12,14,20–42] to estimate GFA. These parameters are calculated using the characteristic temperatures determined by differential thermal analysis (DTA) or differential scanning calorimetry (DSC): the glass transition temperature (Tg), the onset of crystallization (Tx), the crystallization peak (Tc), and the liquidus temperature (Tl).
Explainable Machine Learning Algorithms For Predicting Glass Transition Temperatures
2020, Acta MaterialiaCitation Excerpt :ML has been used in the field of Materials Science and Engineering since the late nineties [9] and has attracted great attention over the last decade. ML algorithms have been used to predict properties of polymers, metallic alloys, and ceramics [8,10–12,14,15,28–35]. In the field of oxide glasses, to the best of our knowledge, the first work to use ANNs was that of Brauer et al. [13], which focused on the prediction of the chemical durability of glasses containing P2O3, CaO, MgO, Na2O, and TiO2.
The role of open spaces to glass-forming ability in bulk metallic glasses
2018, IntermetallicsCitation Excerpt :Metallic glasses with an aperiodic amorphous structure, first discovered in 1960 [1], have stimulated great enthusiasm in their research due to their excellent magnetic properties, high corrosion resistance, and high mechanical properties [2–8]. However, the formation mechanism of bulk metallic glass (BMG) having higher glass-forming ability (GFA) has not been understood yet [9,10] for lack of local structural information inside the amorphous matrix, as e.g., interstitial open spaces. As a typical interstitial open space in the amorphous matrix of BMGs, one can easily speculate a Bernal hole considered in the dense random packing (DRP) model [11] that has been successfully applied to describe the prototype structures of metallic glasses [12–14].