In a recent publication in the journal Nature Synthesis, researchers proposed a robotic artificial intelligence (AI) chemist for autonomous synthesis and optimized catalyst for the oxygen evolution reaction (OER) using Martian meteorites.
Mars has fueled humanity's aspirations for discovering past lives and establishing habitable zones. To achieve sustainable Mars exploration, in situ resource utilization is vital, aiming to reduce mission complexities and costs by harnessing local resources. Crucially, addressing the oxygen supply challenge is paramount, given its pivotal role in propulsion and life support.
Recent indications of water activity on Mars offer an avenue for large-scale oxygen production through solar-powered electrochemical processes. However, synthesizing usable catalysts from local Martian materials poses significant challenges. Overcoming these challenges necessitates an unmanned, self-directing system equipped with AI to efficiently identify optimal catalyst formulas.
Drawing inspiration from advances in robotic chemical synthesis systems, the authors introduce an all-in-one robotic AI chemist. This system, capable of autonomous synthesis and formula design, outpaces conventional trial-and-error approaches by five orders of magnitude.
The proposed AI chemist accelerates the discovery of optimal synthetic formulas, demonstrating its potential for on-site OER electrocatalyst synthesis on Mars and broader applications in planetary exploration. Ongoing developments aim to enhance its adaptability to realistic space conditions, reinforcing its promise for future interstellar exploration.
Optimizing Martian OER catalysts
The study employed high-purity chemicals from Sigma-Aldrich, with the counter and reference electrodes procured from CH Instruments. Deionized water for solutions came from an Ultrapure Water Purification System. For catalyst synthesis, the AI chemist utilized five Martian meteorites, digesting them in an HCl solution to control key catalytic metals. Varying the feedstock proportions in reaction vials allowed for fine adjustments to the metal ratio. This automated procedure was conducted for 243 experiments.
Laser-Induced Breakdown Spectroscopy (LIBS) analyzed meteorites, with a workstation facilitating efficient data collection. A Python program transferred metal molar ratios to Martian ores mass ratios for precise robotic weighing. Mimicking Martian conditions, experiments at -37 °C used aqueous and dry ice solutions. Theoretical calculations employed molecular dynamics (MD) simulations and density functional theory (DFT), with a neural network linking metal composition to catalytic properties optimized via Bayesian methods.
Workflow for OER electrocatalyst synthesis
To streamline the operations of the AI chemist on Mars, a dual-layer workflow is proposed for the on-site synthesis of OER electrocatalysts. The outer layer, managed by a robotic system and 'smart' chemical workstations, conducts a 12-step automated experiment. Simultaneously, the inner layer, orchestrated by an intelligent computational 'brain,' executes nine consecutive digital operations.
Experimental Cycle: Within the experimental cycle, an exploratory robot collects samples of local ore, subjected to LIBS for elemental analysis. The robot conducts chemical and physical pretreatment of ores, including weighing, solution preparation, centrifugation, and solidification. For electrochemical OER testing, the resultant catalyst ink that has been produced with Nafion adhesive serves as the working electrode. Experimental data are transmitted to a cloud server for machine learning processing by the computational 'brain.'
Computational Cycle: In the computational cycle, the 'brain' utilizes MD simulations and DFT calculations to assess the OER activities of high-entropy hydroxides. A theory-based neural network model is trained, incorporating simulation data and robot-driven experimental results. The 'brain' employs a Bayesian algorithm to predict the optimal combination of Martian ores for synthesizing an effective OER catalyst, validated experimentally by the AI chemist.
Machine Learning Models for Overpotential Prediction: The 'brain' creates pre-trained machine learning models, focusing on the OER overpotential as the primary target for optimal catalyst search. Nearly 30,000 unique compositions undergo MD simulations, and the resulting structural features inform the neural network model, achieving accurate reproduction of DFT results. This enables rapid prediction of OER activity by connecting theoretical values with experimentally measured overpotentials. The machine learning model demonstrates exceptional accuracy in predicting true overpotentials.
The AI chemist, armed with LIBS-determined elemental compositions of Martian ores, formulates 243 diverse formulas with randomly selected metal compositions. Electrochemical OER testing yields a range of overpotentials, forming the basis for training a second neural network model. Combining these neural network models enables the anticipation of OER overpotentials across an extensive dataset. This process streamlines Bayesian optimization, facilitating the derivation of the best formula for the desired OER catalyst.
Evaluation: The Kiviat plot illustrates the effectiveness of Bayesian optimization, surpassing experimental-data-guided local searches. The AI-chemist-designed catalyst, with its optimal composition, exhibits superior performance, confirming a substantial improvement over a purely experimental search.
Cycling stability tests demonstrate the catalyst's longevity, sustaining oxygen production for extended periods in various conditions. Under simulated Martian environments, the AI-chemist-designed catalyst exhibits low voltage values and maintains stability, showcasing its potential for on-site oxygen synthesis on Mars. The process's feasibility is affirmed by the catalyst's prolonged stability and estimated oxygen production rates. The AI chemist emerges as a promising tool for advancing OER electrocatalyst synthesis and paving the way for efficient space exploration.
In summary, the research showcases an advanced AI chemist's capability to autonomously synthesize OER catalysts on Mars, covering raw material analysis, pretreatment, synthesis, and testing. In situ optimization efficiently combines experimental and computational data, accelerating model generation and optimal formula identification. The generic, adaptive protocol promises to enhance automated material discovery for extraterrestrial exploration.