Discovering equations that govern experimental materials stability under environmental stress using scientific machine learning

Abstract

While machine learning (ML) in experimental research has demonstrated impressive predictive capabilities, extracting fungible knowledge representations from experimental data remains an elusive task. In this manuscript, we use ML to infer the underlying differential equation (DE) from experimental data of degrading organic-inorganic methylammonium lead iodide (MAPI) perovskite thin films under environmental stressors (elevated temperature, humidity, and light). Using a sparse regression algorithm, we find that the underlying DE governing MAPI degradation across a broad temperature range of 35 to 85 °C is described minimally by a second-order polynomial. This DE corresponds to the Verhulst logistic function, which describes reaction kinetics analogous to self-propagating reactions. We examine the robustness of our conclusions to experimental variance and Gaussian noise and describe the experimental limits within which this methodology can be applied. Our study highlights the promise and challenges associated with ML-aided scientific discovery by demonstrating its application in experimental chemical and materials systems.

Interpretable and Explainable Machine Learning for Materials Science and Chemistry

“…Traditionally, ML models with intrinsic interpretability have proven useful in this setting even if the absence of causal models, for example, in the case of inference and explanation of degradation processes. 34 In this case, the explanations produced by SHAP analysis provided meaningful actionable insights that allowed materials scientists to design new better materials for perovskite solar cells. ML models with a high degree of intrinsic interpretability have been used to identify dominant material descriptors in high dimensional material screening spaces 35 or to actively guide experimental interventions with physical or causal constraints.…”

Iterative symbolic regression for learning transport equations

“…Traditional deterministic algorithms assume a predefined mathematical function and attempt to find parameters with the best fit to the data, whereas, evolutionary algorithms try to find parameters and learn the best‐fit function, simultaneously. Some prevalent methods are genetic programming algorithms, 28–34 sparse regression, 20,35–37 pareto‐optimal regression, 38,39 and the sure‐independence screening and sparsifying operator (SISSO) method 40,41 . Most SR frameworks implement the popular Genetic programming, 42 which is an improved version of Genetic Algorithms (GA), 43,44 inspired by Darwin's theory of natural selection.…”

An Open-Source Environmental Chamber for Materials-Stability Testing Using an Optical Proxy

“…Methods such as 3-dimensional thin plate spline or root-polynomial regression are applied to convert the colors seen by the camera to the colors they would look like under a standardized reference illuminant using a reference color chart as a basis for the transformation [33], [34]. Calibrated color has been previously introduced as a quantitative metric to evaluate dye solar cell degradation [35], [36] and is used especially in color analysis applications that involve multiple devices or sites. Furthermore, calibrated color is useful when small-area colorimeters are not sufficient due to the spatial variability of the samples such as in the food industry [27], [29], [37].…”