Estimating the performance of a material in its service space via Bayesian active learning: a case study of the damping capacity of Mg alloys

Abstract

In addition to being determined by its chemical composition and processing conditions, the performance of a material is also affected by the variables of its service space, including temperature, pressure, and frequency. A rapid means to estimate the performance of a material in its service space is urgently required to accelerate the screening of materials with targeted performance. In the present study, a materials informatics approach is proposed to rapidly predict performance within a service space based on existing data. We utilize an active learning loop, which employs an ensemble machine learning method to predict the performance, followed by a Bayesian experimental design to minimize the number of experiments for refinement and validation. This approach is demonstrated by predicting the damping properties of a ZE62 magnesium alloy in a service space defined by frequency, strain amplitude, and temperature based on the available data for other magnesium alloys. Several utility functions that recommend a particular experiment to refine the estimates of the service space are used and compared. In particular, the standard deviation is found to reduce the prediction error most efficiently. After augmenting the database with nine new experimental measurements, the uncertainties associated with the predicted damping capacities are largely reduced. Our method allows us to forecast the properties in the service space of a given material by rapid refinement of the predictions via experiment measurements.

Searching new cocrystal structures of CL-20 and HMX via evolutionary algorithm and machine learning potential

“…The results demonstrate a good agreement between our ReaxFF-nn potential model and the DFT calculations. Although active learning algorithms [29,61,62] to collect data automatically are available, the application to molecular systems requires further tests, and we also plan to develop an uncertainty-driven active learning algorithm based on the Z-matrix.…”

An active machine learning approach for optimal design of magnesium alloys using Bayesian optimisation

“…In particular, so-called active machine learning approaches, which combine human expertise with iterative model refinement, have demonstrated great potential in reducing the experimental burden and maximising the search efficiency in materials design 11 – 14 . Bayesian optimisation and adaptive design are methods following an active ML strategy, which require goal-directed iterative feedback 6 , 15 , 16 .…”