Estimating “depth of layer” (DOL) in ion-exchanged glasses using explainable machine learning

The process of ion exchange includes the substitution of a larger ion for a smaller ion, which frequently takes place at temperatures lower than the glass transition temperature. The crucial variables in this particular procedure are the depth of layer (DOL) and the surface compressive stress (CS). Determining the best composition for ion-exchangeable glasses is an intricate task that is impeded by conflicting demands. Traditional experimental and empirical discovery methods are challenging, expensive, and time-consuming. This paper presents a machine learning model to understand the effects of many factors, such as chemical compositions, temperature, process time, ion source, and electrical field strength, on the DOL. A dataset of ion-exchangeable glasses is acquired by extracting the data from academic research articles and patents. The proposed dataset contains 265 unique samples, each characterized by 22 parameters that are employed as learning features. A meta-estimator called a voting regressor, which performs better than individual models, is presented using the three best-performing models as inputs. The R2 score of the DOL prediction model is 0.8841. The proposed model was interpreted by computing the Shapley additive explanations (SHAP) values of the ion exchange process variables as well as the chemical composition of the glass. The enhancement of the DOL value was found to be highly influenced by the ion exchange process parameters, such as duration, temperature, and electric field intensity, as well as the compositions of Si2O, Na2O, and Li2O. In contrast, the DOL value decreases with an increase in Tg, Al2O3, MgO, CaO, K2O, TiO2, and B2O3.