HYDERABAD: Carbon dioxide is usually the problem, but Hyderabad scientists have developed a machine learning tool that could aid in the process of turning it into dimethyl ether (DME), a fuel that can be used as a diesel alternative or blended with LPG. The tool can predict which catalyst combinations are most likely to work, potentially reducing years of trial-and-error in laboratories.
Researchers at CSIR-Indian Institute of Chemical Technology (IICT) have developed a machine learning framework to speed up the search for catalysts that can convert carbon dioxide into DME. The framework predicts how different catalyst compositions and operating conditions could affect the amount of carbon dioxide converted and the quantity of DME produced.
The study was authored by Ganesh Kumar Ramachandran, Banoth Upendar, Reddi Kamesh, Ashok Jangam, Sreepriya Vedantam and Venugopal Akula of CSIR-IICT. For the study, researchers compiled 330 experimental results from 39 peer-reviewed studies. They used 16 factors, including catalyst characteristics and reaction conditions, to train and test several machine learning models.
A Gradient Boosted Regression Tree (GBRT) model performed best. When tested on data it had not previously seen, it achieved scores of 0.92 for predicting carbon dioxide conversion and 0.94 for predicting DME selectivity, indicating a high level of predictive accuracy.
The researchers found that reaction temperature, pressure and the Si/Al ratio of the acid catalyst were among the most important factors affecting performance.
Importantly, the model uses information available before a catalyst is made, such as the elemental composition of active and promoter metals and properties of the catalyst support. This means researchers could potentially screen promising catalyst combinations on a computer before making them in a laboratory, reducing the number of experiments required.
DME, also known as methoxymethane, can be made from captured carbon dioxide and hydrogen. It can be used as a transportation fuel, blended with LPG and in chemical manufacturing.
In India, its use has largely been limited to pharmaceutical and specialty industries, with significant quantities imported from China and Japan. The study also points to the potential of blending DME with LPG. Even a 20% DME blend could have a significant impact given India’s large LPG consumer base and dependence on imports.
However, converting carbon dioxide directly into DME remains difficult. The process combines two steps — producing methanol from carbon dioxide and then converting methanol into DME.
Carbon dioxide is chemically stable and difficult to activate, while competing reactions, catalyst degradation and water formation can reduce efficiency.
The researchers said the model could eventually help optimise catalysts and design new ones. However, predictions would still need to be tested rigorously in laboratories before any industrial application.