AIPressRoom’s Post

Machine Learning Algorithm to Predict Methane Adsorption Capacity of Coal | Energy & Fuels: By training the model using large data sets, issues of error and reproducibility in traditional experiments are addressed, improving experimental ... #bigdata #cdo #cto

With methane’s storage difficulties, covalent organic frameworks (COFs) have been discussed as an alternative. However identifying potential COFs is an extensive process that is limiting progress on methane storage alternatives. Researchers in ME have developed a new approach that combines machine learning with symbolic regression, resulting in easily-interpretable equations that can accurately predict methane storage capacity between each COF. “By prioritizing physical, meaningful and measurable features, we’ve made it easier for experimentalists to apply these models directly, enabling broader participation in the field and accelerating the development of high-performance materials,” said ME associate research scientist and corresponding author of the study Alauddin Ahmed. Learn more about the hundreds of COFs the team has identified so far with their new model [article] https://bit.ly/4fXczWO

I am excited to share that our latest research, titled "Physics-Based Machine Learning Models Predict Carbon Dioxide Solubility in Chemically Reactive Deep Eutectic Solvents," has been published in the ACS Omega. In this work, we've developed four cutting-edge machine learning (ML) models to predict the solubility of carbon dioxide (CO2) in chemically reactive deep eutectic solvents (DESs). What sets our approach apart is the incorporation of physics-informed input features, enabling us to capture accurate intra and intermolecular interactions without the complexities of directly simulating DES-CO2 systems and their chemical reactions. Furthermore, we have leveraged our newly developed ML model for predicting DES viscosity and use it as an additional input in our CO2 solubility predictions. This research is a significant step forward as we compile a comprehensive dataset of CO2 solubility in chemically reactive DESs, covering a wide range of chemical spaces of both hydrogen bond acceptors (HBA) and hydrogen bond donors (HBD). Our models not only achieve impressive accuracy and accelerate CO2 capture research but also guide the researchers to understand the interactions and key features that are important for the potential design of chemically reactive DESs for CO2 capture. Check out the full article here: https://lnkd.in/e-azx4Zj Viscosity Paper: https://lnkd.in/eQqU5yWj Thanks to the team, Michelle K. Kidder Omar Demerdash Blake Simmons @Seema Singh, and Jeremy Christopher Smith, for their guidance and contributions 🙌. Feel free to reach out for further discussions or collaborations. Machine Learning #computationalchemistry #deepeutecticsolvents #greensolvents #sustainability #carboncapture

Researchers used machine learning to predict and enhance the specific capacitance of N-doped porous biochar electrodes for supercapacitors. The Random Forest model proved most accurate, revealing that optimizing pore structure is more crucial than nitrogen doping, offering insights for better biochar electrode design. #Biochar #Pyrolysis #CarbonCapture

🔬 Innovative Approach: Machine Learning Helps Overcome a Key Hydrogen Economy Challenge As we race towards sustainable energy solutions, hydrogen emerges as a promising carrier. But there's a catch - hydrogen embrittlement (HE) poses a serious threat to metal infrastructure. Now, researchers from NTNU and SINTEF have developed an innovative ML approach to predict how the material behaves under hydrogen exposure. Key highlights from their groundbreaking research: 🎯 The team created a Gradient Boosting Machine model achieving 88.6% accuracy in predicting material susceptibility to hydrogen embrittlement. 🔋 This advancement could revolutionize how we select materials for hydrogen infrastructure, from storage tanks to fuel cell components. 💡 The model considers multiple factors simultaneously: environmental conditions, material properties, and loading conditions - something traditional methods struggle with. Why this matters: Understanding material behavior becomes crucial for safety and reliability as we transition to hydrogen-based technologies. This ML approach could significantly reduce testing time and costs while improving our ability to predict potential failures. What's next? The researchers suggest expanding the database and including dynamic load conditions to enhance prediction accuracy further. This could be a game-changer for the hydrogen industry's safety standards. Link: https://lnkd.in/d2mu_YVt #MachineLearning #MaterialsScience #HydrogenEnergy #HydrogenSafety #EnergyTransition

Understanding the oxidation states of ions is crucial for inferring structure-property relationships in materials. This new electrochemical series should make that a bit easier. https://lnkd.in/gzZt4Ezq

Carnegie Mellon University and Los Alamos National Laboratory have just unlocked a new horizon in chemistry with their machine learning model, ANI-1xnr. This promises to drastically reduce the computational power and time needed for simulating diverse organic reactions. Imagine exploring the vast possibilities in drug discovery, biofuel analysis, and beyond, all thanks to machine learning. Read more: #ChemicalProcessing #OrganicReactionsSimulations  

Machine learning exploration of experimental conditions for optimized electrochemical CO₂ reduction Optimizing the electrochemical reduction of CO₂ (CO₂RR) is a critical step in advancing sustainable technologies for carbon management. However, identifying the optimal conditions for achieving high product selectivity and efficiency often requires extensive experimental effort. Recent work by Setyowati et al. demonstrates how machine learning (ML) can significantly streamline this process. The study focuses on using ML to predict the outcomes of CO₂RR under varying experimental conditions, employing a dataset derived entirely from experimental results. Parameters such as Ag catalyst density (Ag-dens), the Ag:Nafion ratio in catalyst inks, and cathodic current density (CCD) were key inputs for the model. Advanced ML techniques, including Random Forest Regression and Kernel Ridge Regression, were applied to predict faradaic efficiency for CO (FECO) and the ratio of CO to H₂ (ξ), achieving cross-validated mean absolute errors of 8.18% and 5.58%, respectively. A major innovation of this study is its use of inverse analysis to propose experimental conditions that maximize FECO or achieve specific CO/H₂ ratios. By generating synthetic datasets and employing principal component analysis (PCA), the model identified optimal conditions that were experimentally validated, confirming its predictive power. These validations showed that the ML-driven approach could achieve desired outcomes with significantly fewer experimental trials. Paper (open access): https://lnkd.in/dmCyxk95 #MachineLearning #SustainableTech #Electrochemistry #CO2Reduction #GreenTechnology #DataDrivenResearch #CatalystDesign #CarbonNeutral #MaterialScience #CleanEnergy #AIForScience #EnergyInnovation #ClimateSolutions #AdvancedMaterials #MLApplications

AlvaDesc has been used in Pandey, S. K., & Kunal Roy (2024). Predicting the performance and stability parameters of energetic materials (EMs) using a machine learning-based q-RASPR approach. Energy Advances. Abstract: The performance and stability are the two major areas of concern related to energetic materials (EMs). Balancing both the performance and stability simultaneously can result in the development of new advanced compounds that will not only perform better but at the same time be highly stable to physical/chemical/thermal stress. In this study, we aimed to predict some of the properties related to detonation performance (density, n = 12 805; gas-phase heat of formation, n = 2565) and thermal stability (decomposition temperature, n = 656; melting point, n = 19 667) of EMs using the quantitative Read-Across Structure–Property Relationship (q-RASPR) approach. q-RASPR, a combined application of quantitative structure–property relationship (QSPR) and RA methodologies, has shown an enhancement in the model predictivity, compared to the traditional QSPR method. The data sets collected from var- ious sources were first curated to prepare high-quality data. After the structural representation of the data points and descriptor calculation, each data set was divided into the respective training and test sets. Different methodologies were employed to train the model, and the models so developed were validated based on the Organization for Economic Cooperation and Development (OECD) principles. Also, the developed models’ predictivity was checked using different ML algorithms. All the developed models showed good statistical quality with R2 values (training set) ranging from 0.64 for decomposition temperature and 0.75 for the melting point to 0.94 for density and heat of formation data sets. Also, the external validation results were quite promising, which indicates that the predictive power of our developed models was significant. The models so developed can be used for examining the performance and heat resistance capacity of the newly developed compounds, screening of databases, modification of older derivatives, and/or the development of heat-resistant (non-thermo-labile) and impactful EMs. You can find more information on alvaDesc at: https://lnkd.in/gyaHaaM #researchpaper #alvadesc #moleculardescriptors #machinelearning #qsar #cheminformatics #compchem #insilico

'the research team devlpd a set of machine-learng (ML) models known as "feedforward neural nets" to screen 50,000 polymers for optimal properties, includg ability to withstand hi-temps & strong electric fields, hi-energy storag density & ease of synthesis' ~ #filmCapacitors #ML #machineLearning #FeedforwardNeuralNetworks #polymers #electricfields #energyStorage #BerkeleyLab’sMolecularFoundry #NatureEnergy #njTare