AI-designed hierarchical SAPO-34 catalyst breakthrough with experimental verification†
Abstract
The reliance on resource-intensive empirical methods has traditionally hindered the development of heterogeneous catalysts. The integration of machine learning offers a paradigm shift, enabling accelerated discovery and deeper insights into the formation of complex catalyst structures. Employing an artificial intelligence approach, this study introduces the first-ever prediction of key properties of SAPO-34, including BET, micropore, and external surface area, as well as average particle size. Two machine learning models, artificial neural network and random forest, were employed alongside statistical techniques for in-depth analysis of the extracted dataset, while a SAPO-34 catalyst was synthesized and characterized using XRD, BET, and SEM to verify the predictions. Sensitivity analysis revealed the most influential synthesis parameters for each property. Namely, crystallization time for BET surface area, Al2O3 content for micropore surface area, precursor sequence addition (PSAd) for external surface area, and TEAOH concentration for particle size were found to regulate the physicochemical properties of the developed catalyst. The RF models showed high accuracy with R2 values of 0.8765, 0.8894, and 0.9698 for BET, micropore, and external surface areas, respectively, while the ANN model excelled in predicting average particle size with an R2 of 0.9950. Experimental verification confirmed the minimal errors, with predictions of BET and micropore surface areas deviating by just 3% and 4%. This work paves the way for optimizing SAPO-34 synthesis, advancing its industrial application, and providing key insights into AI-driven catalyst design.