machine learning prediction of flow stress of bulk metallic glasses during thermoplastic deformation within the supercooled liquid region

Bulk metallic glasses (bmgs) have attracted considerable interest over the past two decades because of their exceptional properties and unique structure. bmgs exhibit superplastic deformability in the supercooled liquid region (slr). bmgs can be thermoplastically deformed into complex-shaped components in a single step. as a result of their thermoplastic forming ability, bmgs can be used in a wide range of applications, including microelectronic mechanical systems (mems). cuzr-based bmgs have attracted great attention from scholars due to their excellent properties, such as high strength, high hardness, high wear resistance, small thermal expansion coefficient and excellent thermoplastic properties, etc. a bmg is characterized structurally by its unique atomic structure combining short-range order with long-range disorder. the short-range order can be addressed as clusters with an atom at the center and some shell atoms. although there is no dislocation accompanying the deformation of an amorphous structure, it is believed that free volumes between atoms play a crucial role during the deformation process. in short, it can be inferred that applied stress during deformation is highly correlated with free volumes through the atomic structure. the thermoplastic flow of bmgs is usually examined by hot compression test in slr. workability, flow stress and stress overshoot are the most important results that can be obtained by hot compression test in slr. although there have been many investigations on the thermoplastic flow of bmgs, it is not quite simple to design an appropriate bmgs that fits the thermoplastic deformation process. the complexity of parameters affecting properties of a metallic glass leads to trail-and-error to develop a novel appropriate bmg. computational approach, especially data driven approach has been used as a tool to predict the properties of bulk metallic glasses and assist the process of alloy design. in this work, a data-driven approach was proposed to predict the flow stress of cuzr-based bmgs during thermoplastic deformation in supercooled liquid region. the process of developing a machine learning (ml) model to predict the flow stress consists of 3 steps including data collection, data processing and feature calculation and the third step is training ml algorithms using the data. first, data from literature and digital resources was collected. the dataset in this work consists of 14 alloys which were experimentally subjected to hot compression test and the results of tests were used to prepare the dataset. the whole dataset contains 348 lines of data of hot compression test of bmgs within slr. second, the data mined from literature were processed and features were calculated. in this work, mixing enthalpy, mixing entropy, minimum self-diffusion of elements from each alloy as the diffusion control element, strain rate and (t- tg)/(tg- tx ) as features were used to train ml models. before employing any ml algorithm, dataset was split to a training set and a test set. in order to develop a reliable and robust model, feature-wised standard normalization as a preprocessing on the dataset was applied. training set was used to train ml algorithms like k-nearest neighbor, random forest, xgboost. many grid searches, k-fold cross validations were performed to tune the hyperparameters, in order to get the best algorithm. our results show that the xgboost algorithm can perform very well (r2=86%) to predict flow stress during thermoplastic deformation. the most important features are (t-tg)/(tg-tx) and strain rate which is consistent with previous studies.