Electrocaloric property optimization of PMN-PT ceramics using predictive modeling

Abstract

An eXtreme Gradient Boosting (XGBoost) machine learning model was developed to predict the electrocaloric temperature change (Delta TEC) of PMN-PT electroceramics based on processing parameters (PMN/PT ratio, sample type, calcination and sintering time/temperature) and measurement conditions (direct/indirect measurement, electric field, and measurement temperature). A dataset of 2863 data points was compiled from the experimental literature, and the model achieved a coefficient of determination (R2) of 0.97 and a mean absolute error (MAE) of 0.04 degrees C on the test set, demonstrating strong predictive performance. Feature importance analysis revealed that electric field and T-TC are the primary drivers of electrocaloric response, while processing conditions also significantly influence Delta TEC. A Bayesian Optimization framework was employed to systematically identify the optimal processing parameters for maximizing Delta TEC. Under experimentally realistic conditions (24 degrees C measurement temperature, 40 kV/cm electric field), the optimal PMN/PT composition (81.06%/18.94%) and processing parameters (calcination at 810 degrees C for 3 h, sintering at 1280 degrees C for 7 h) were identified, yielding a predicted Delta TEC of 1.40 degrees C, which exceeds most experimentally reported values under similar conditions. Additionally, a virtual dataset of 100.000 synthetic data points was generated and analyzed to explore high-performance regions within the electrocaloric design space. The results highlight the potential of machine learning in materials science, offering a scalable and data-driven approach to accelerate the discovery and optimization of advanced electrocaloric materials.