skip to main content


Title: Machine learning enhanced spectroscopic analysis: towards autonomous chemical mixture characterization for rapid process optimization
Autonomous chemical process development and optimization methods use algorithms to explore the operating parameter space based on feedback from experimentally determined exit stream compositions. Measuring the compositions of multicomponent streams is challenging, requiring multiple analytical techniques to differentiate between similar chemical components in the mixture and determine their concentration. Herein, we describe a universal analytical methodology based on multitarget regression machine learning (ML) models to rapidly determine chemical mixtures' compositions from Fourier transform infrared (FTIR) absorption spectra. Specifically, we used simulated FTIR spectra for up to 6 components in water and tested seven different ML algorithms to develop the methodology. All algorithms resulted in regression models with mean absolute errors (MAE) between 0–0.27 wt%. We validated the methodology with experimental data obtained on mixtures prepared using a network of programmable pumps in line with an FTIR transmission flow cell. ML models were trained using experimental data and evaluated for mixtures of up to 4-components with similar chemical structures, including alcohols ( i.e. , glycerol, isopropanol, and 1-butanol) and nitriles ( i.e. , acrylonitrile, adiponitrile, and propionitrile). Linear regression models predicted concentrations with coefficients of determination, R 2 , between 0.955 and 0.986, while artificial neural network models showed a slightly lower accuracy, with R 2 between 0.854 and 0.977. These R 2 correspond to MAEs of 0.28–0.52 wt% for mixtures with component concentrations between 4–10 wt%. Thus, we demonstrate that ML models can accurately determine the compositions of multicomponent mixtures of similar species, enhancing spectroscopic chemical quantification for use in autonomous, fast process development and optimization.  more » « less
Award ID(s):
1943972
NSF-PAR ID:
10319787
Author(s) / Creator(s):
; ; ;
Date Published:
Journal Name:
Digital Discovery
Volume:
1
Issue:
1
ISSN:
2635-098X
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Fugacity is a fundamental thermodynamical property of gas and gas mixtures to determine their behavior and dynamics in complex systems. Fugacity can be deduced experimentally from the measurements of volume as a function of pressure at constant temperature or calculated iteratively using analytical equations of states (EOS). Experimental measurement is time-consuming, and analytical models based on EOS are computationally demanding, especially when an approximate but quick estimation is desired. In this work, machine learning (ML) is employed as a viable alternative to analytical EOSs for quick and accurate approximation of CO2 fugacity coefficients. Five different ML algorithms are used to estimate the fugacity coefficients of pure CO2 as a function of pressure (≤ 2000 bar) and temperature (≤ 1000 °C). A combination of experimental and pseudo-experimental (obtained from an analytical EOS) data of CO2 fugacity coefficients is used to train, validate, and test the models. The best results were found using the Extreme Gradient Boosting algorithm, which showed a mean square error of only 0.0002 in the validation data and an average deviation of only 1.3% in the test data (pure prediction). To quantify the effectiveness of the machine learning techniques, results from the best-performing model are compared with two state-of-the-art analytical models. The ML model with significantly less computational complexity showed similar accuracy to the analytical models. The estimated fugacity data are then used to compute the CO2 solubility in aqueous NaCl solution of different concentrations, and a maximum deviation of only 3.2% from the experimental data is observed. 
    more » « less
  2. Tire wear particles (TWPs) are a major category of microplastic pollution produced by friction between tires and road surfaces. This non-exhaust particulate matter (PM) is transported through the air and with runoff leading to environmental pollution and health concerns. Here, we collected airborne PM along paved roads with different traffic volumes and speeds using Sigma-2 passive samplers. Particles entering the samplers deposit onto substrates for analysis, or, as we modified it, directly into small (60 ml) separatory funnels, which is particularly useful with high particle loads, where a density separation aids in isolating the microplastics. We quantified putative TWPs (∼10–80 µm) deposited on the substrates (primarily adhesive tape on glass slides) and in the funnels using stereomicroscopy. Putative TWP deposition rates (particles/cm 2 /day ± SD) at 5 m from the road were highest near a busy highway (324 ± 129), followed by a boulevard with moderate traffic (184 ± 93), and a slow traffic avenue (29 ± 7). We observed that deposition rates increased within proximity to the highway: 99 ± 54, 180 ± 88, and 340 ± 145 at 30, 15, and 5 m, respectively. We show that TWP abundances (i.e., deposition and mass concentration) increase with vehicle braking (driving behavior). We observed no differences ( p > 0.05) between the separatory funnel and adhesive tape collection methods. In addition, we were able to obtain FTIR spectra of TWPs (>10 µm) using µ-ATR-FTIR. Both deserve further scrutiny as novel sampling and analytical approaches. In a separate sampling campaign, we differentiated 1438 particles (∼1–80 µm) deposited on boron substrates into TWP, metal, mineral, and biogenic/organic classes with single particle SEM/EDX analysis based on morpho-textural-chemical classification and machine learning. The results revealed similar concentration trends with traffic (high > moderate > low), with the distribution of particle sources alike for the highway and the moderate road: TWPs (∼38–39%) > biogenic (∼34–35%) > minerals (∼23–26%), and metallic particles (∼2–3%). The low traffic road yielded a much different distribution: biogenic (65%) > minerals (27%) > TWPs (7%) > metallic particles (1%). Overall, this work provides much-needed empirical data on airborne TWPs along different types of roads. 
    more » « less
  3. null (Ed.)
    The di!usion of water through silicate melts is a key process in volcanic systems. Di!usion controls the growth of the bub- bles that drive volcanic eruptions and determines the evolution of the spatial distribution of dissolved water during and after magma mingling, crystal growth, fracturing and fragmentation, and welding of pyroclasts. Accurate models for water di!u- sion are therefore essential for forward modelling of eruptive behaviour, and for inverse modelling to reconstruct eruptive and post-eruptive history from the spatial distribution of water in eruptive products. Existing models do not include the kinetics of the homogeneous species reaction that interconverts molecular (H2Om) and hydroxyl (OH) water; reaction kinetics are impor- tant because final species distribution depends on cooling history. Here we develop a flexible 1D numerical model for di!usion and speciation of water in silicate melts. We validate the model against FTIR transects of the spatial distribution of molecular, hydroxyl, and total water across di!usion-couple experiments of haplogranite composition, run at 800–1200 C and 5 kbar. We adopt a stepwise approach to analysing and modelling the data. First, we use the analytical Sauer-Freise method to deter- mine the e!ective di!usivity of total water DH2Ot as a function of dissolved water concentration CH2Ot and temperature T for each experiment and find that the dependence of DH2 Ot on CH2 Ot is linear for CH2 Ot K 1:8 wt.% and exponential for CH2 Ot J 1:8 wt.%. Second, we develop a 1D numerical forward model, using the method of lines, to determine a piece-wise function for DH2 Ot !CH2 Ot ; T " that is globally-minimized against the entire experimental dataset. Third, we extend this numerical model to account for speciation of water and determine globally-minimized functions for di!usivity of molecular water DH2 Om !CH2 Ot ; T " and the equilibrium constant K for the speciation reaction. Our approach includes three key novelties: (1) functions for dif- fusivities of H2Ot and H2Om, and the speciation reaction, are minimized simultaneously against a large experimental dataset, covering a wide range of water concentration (0:25 CH2 Ot 7 wt.%) and temperature (800  C T 1200  C), such that the resulting functions are both mutually-consistent and broadly applicable; (2) the minimization allows rigorous and robust analysis of uncertainties such that the accuracy of the functions is quantified; (3) the model can be straightforwardly used to determine functions for di!usivity and speciation for other melt compositions pending suitable di!usion-couple experiments. The modelling approach is suitable for both forward and inverse modelling of di!usion processes in silicate melts; the model is available as a MATLAB script from the electronic supplementary material. 
    more » « less
  4. null (Ed.)
    The di!usion of water through silicate melts is a key process in volcanic systems. Di!usion controls the growth of the bub- bles that drive volcanic eruptions and determines the evolution of the spatial distribution of dissolved water during and after magma mingling, crystal growth, fracturing and fragmentation, and welding of pyroclasts. Accurate models for water di!u- sion are therefore essential for forward modelling of eruptive behaviour, and for inverse modelling to reconstruct eruptive and post-eruptive history from the spatial distribution of water in eruptive products. Existing models do not include the kinetics of the homogeneous species reaction that interconverts molecular (H2Om) and hydroxyl (OH) water; reaction kinetics are impor- tant because final species distribution depends on cooling history. Here we develop a flexible 1D numerical model for di!usion and speciation of water in silicate melts. We validate the model against FTIR transects of the spatial distribution of molecular, hydroxyl, and total water across di!usion-couple experiments of haplogranite composition, run at 800–1200 C and 5 kbar. We adopt a stepwise approach to analysing and modelling the data. First, we use the analytical Sauer-Freise method to deter- mine the e!ective di!usivity of total water DH2Ot as a function of dissolved water concentration CH2Ot and temperature T for each experiment and find that the dependence of DH2 Ot on CH2 Ot is linear for CH2 Ot K 1:8 wt.% and exponential for CH2 Ot J 1:8 wt.%. Second, we develop a 1D numerical forward model, using the method of lines, to determine a piece-wise function for DH2 Ot !CH2 Ot ; T " that is globally-minimized against the entire experimental dataset. Third, we extend this numerical model to account for speciation of water and determine globally-minimized functions for di!usivity of molecular water DH2 Om !CH2 Ot ; T " and the equilibrium constant K for the speciation reaction. Our approach includes three key novelties: (1) functions for dif- fusivities of H2Ot and H2Om, and the speciation reaction, are minimized simultaneously against a large experimental dataset, covering a wide range of water concentration (0:25 CH2 Ot 7 wt.%) and temperature (800  C T 1200  C), such that the resulting functions are both mutually-consistent and broadly applicable; (2) the minimization allows rigorous and robust analysis of uncertainties such that the accuracy of the functions is quantified; (3) the model can be straightforwardly used to determine functions for di!usivity and speciation for other melt compositions pending suitable di!usion-couple experiments. The modelling approach is suitable for both forward and inverse modelling of di!usion processes in silicate melts; the model is available as a MATLAB script from the electronic supplementary material. 
    more » « less
  5. The granitic water-saturated solidus (G-WSS) is the lower temperature limit of magmatic mineral crystallization. The accepted water-saturated solidus for granitic compositions was largely determined >60 years ago1. More recent advances in experimental petrology, improved analytical techniques, and recent observations that granitic systems can remain active or spend a significant proportion of their lives at conditions below the traditional G-WSS2–5 necessitate a careful experimental investigation of the near-solidus regions of granitic systems. Natural and synthetic starting materials were melted at 10 kbar and 900°C with 48 wt% H2O to produce hydrous glasses for subsequent experiments at lower PT conditions used to locate the G-WSS. We performed crystallization experiments and melting experiments at temperatures ranging from 575 to 800°C and 1, 6, 8, and 10 kbar on 12 granitoid compositions. First, we ran a series of isothermal crystallization experiments along each isobar at progressively lower temperatures until runs completely crystallized to identify apparent solidus temperatures. Geochemical analyses of quenched glass compositions demonstrate that progressive crystallization drives all starting compositions towards silica-rich, water-saturated rhyolitic/granitic melts (e.g., ~7578 wt% SiO2). After identifying the apparent solidus temperatures at which the various compositions crystallized, we then ran series of reversal-type melting experiments. With the goal of producing rocks with hydrous equilibrium microstructures, we crystallized compositions at temperatures ~10°C below the apparent solidus identified in crystallization experiments, and then heated isobarically to conditions that produced ~20% melt during the crystallization experiments. Importantly, crystallization experiments and heating experiments at the same PT conditions produced similar proportions of melt, crystals, and vapor. A time-series of experiments 230 days at PT conditions previously identified to produce ~10% to 20% melt did not reveal any kinetic effects on melt crystallization. Experiments at 6 to 10 kbar crystallized/melted at temperatures close to the published G-WSS. However, at lower pressures where the published G-WSS is strongly curved in PT space, all compositions investigated contained melt to temperatures ~75 to 100°C below the accepted G-WSS. The similarity of crystallization temperatures for the higher-pressure experiments to previously published results, similar phase proportions in melting and crystallization experiments, and the lack of kinetic effects on crystallization collectively suggest that our lower pressure constraints on the G-WSS are accurate. The new experimental results demonstrating that the lower-pressure G-WSS is significantly lower than unanimously accepted estimates will help us to better understand the storage conditions, evolution, and potential for eruption in mid- to upper-crustal silicic magmatic systems. (1) Tuttle, O.; Bowen, N. Origin of Granite in the Light of Experimental Studies in the System NaAlSi3O8–KAlSi3O8–SiO2–H2O; Geological Society of America Memoirs; Geological Society of America, 1958; Vol. 74. https://doi.org/10.1130/MEM74. (2) Rubin, A. E.; Cooper, K. M.; Till, C. B.; Kent, A. J. R.; Costa, F.; Bose, M.; Gravley, D.; Deering, C.; Cole, J. Rapid Cooling and Cold Storage in a Silicic Magma Reservoir Recorded in Individual Crystals. Science 2017, 356 (6343), 1154–1156. https://doi.org/10.1126/science.aam8720. (3) Andersen, N. L.; Jicha, B. R.; Singer, B. S.; Hildreth, W. Incremental Heating of Bishop Tuff Sanidine Reveals Preeruptive Radiogenic Ar and Rapid Remobilization from Cold Storage. Proceedings of the National Academy of Sciences 2017, 114 (47), 12407–12412. https://doi.org/10.1073/pnas.1709581114. (4) Ackerson, M. R.; Mysen, B. O.; Tailby, N. D.; Watson, E. B. Low-Temperature Crystallization of Granites and the Implications for Crustal Magmatism. Nature 2018, 559 (7712), 94–97. https://doi.org/10.1038/s41586-018-0264-2. (5) Glazner, A. F.; Bartley, J. M.; Coleman, D. S.; Lindgren, K. Aplite Diking and Infiltration: A Differentiation Mechanism Restricted to Plutonic Rocks. Contributions to Mineralogy and Petrology 2020, 175 (4). https://doi.org/10.1007/s00410-020-01677-1. 
    more » « less