Guided electrocatalyst design through in-situ techniques and data mining approaches

Abstract

Intuitive design strategies, primarily based on literature research and trial-and-error efforts, have significantly contributed to advancements in the electrocatalyst field. However, the inherently time-consuming and inconsistent nature of these methods presents substantial challenges in accelerating the discovery of high-performance electrocatalysts. To this end, guided design approaches, including in-situ experimental techniques and data mining, have emerged as powerful catalyst design and optimization tools. The former offers valuable insights into the reaction mechanisms, while the latter identifies patterns within large catalyst databases. In this review, we first present the examples using in-situ experimental techniques, emphasizing a detailed analysis of their strengths and limitations. Then, we explore advancements in data-mining-driven catalyst development, highlighting how data-driven approaches complement experimental methods to accelerate the discovery and optimization of high-performance catalysts. Finally, we discuss the current challenges and possible solutions for guided catalyst design. This review aims to provide a comprehensive understanding of current methodologies and inspire future innovations in electrocatalytic research.

Keywords: Catalytic mechanism; Data mining; In-situ experimental techniques; Mechanism guidance; Structural-property relationship.

Fig. 1

Atomic-scale identification of active sites with SECM and SECCM techniques. (A-B) Schematic representation of (A) positive feedback produced by oxidation/reduction of ferrocene methanol (Fc) and (B) substrate generation/tip collection of dioxygen at NiO nanosheet. (C-D) SECM imaging of NiO nanosheet with the HOPG as the substrate at the (C) Feedback mode and (D) SG/TC mode. Reproduced with permission from [37]. Copyright 2019, National Academy of Sciences. (E) High-angle annular dark-field scanning transmission electron microscopy (HAADF STEM) image of PPy-CuPcTs (40:1). (F) Atom utilization and active metal content based on SI-SECM titration and ICP-MS results. Reproduced with permission from [39]. Copyright 2020, Elsevie. (G) Atomic force microscopy (AFM) image of the designed setup of a 2-μm-diameter circular aperture in the SiN_x substrate covered with monolayer graphene. (H) Schematic of SECCM setup. (I) Example of SECCM maps (I_collector maps) for apertures covered with graphene. Reproduced with permission from [28]. Copyright 2023, Springer Nature. (J-K) SECCM current maps of (J) 2D MoS₂ and (K) 2D MoS₂ after electrochemical activation on HOPG substrate. Reproduced with permission from [46]. Copyright 2020, Wiley

Fig. 2

Identification of active sites with the SMF imaging. (A) Schematic representation of the experimental setup used for imaging H₂ nanobubbles during electrocatalytic water splitting. (B) TIRF image at the applied voltage of -1.8 V. (C) Scatter plots depicting the accumulated spatial distribution of H₂ nanobubbles observed during the potential scan from − 1.5 V to − 1.8 V. Reproduced with permission from [32]. Copyright 2018, National Academy of Sciences. (D) Schematic diagram of the on-chip TIRF setup. (E) Overlay of local activity and AFM topology of monolayer MoS₂ with a high density of protrusions. (F) The plot of local activity represented by nanobubble density as the function of tensile (blue) and compressive (red) strain. (G) AFM topology of a typical single protrusion. Note that the dotted polygonal rings denote the boundaries of radial segments (width: 39 nm). (H) Overlay image of strain map and nanobubbles (purple). (I) Averaged nanobubble density versus strain value across the white dotted line in panel H. Reproduced with permission from [58]. Copyright 2024, American Chemical Society

Fig. 3

Atomic-scale visualization of active sites using EC-STM technique. (A-B) Schematic diagram of EC-STM principles, showing that the amplitude of the tunneling-current noise is proportional to the reactivity of surface sites. (C) EC-STM line scans (constant-current mode) obtained over a Pt(111) surface in 0.1 M HClO₄. Reproduced with permission from [26]. Copyright 2017, Springer Nature. (D) EC-STM topographic images of the Gr/Fe (1.8monolayer)/Pt(111) surface recorded at E = 195 mV. STM image showing the different structural units investigated, Fe-2 V, Fe-3 V, Fe-4 V, 3Fe-6 V. (E) Normalized current roughness, L, as a function of E extracted from the areas outlined by the rectangular boxes in D. Reproduced with permission from [10]. Copyright 2021, Springer Nature

Fig. 4

Investigation of reaction mechanism using the operando XAS and XPS. (A) Schematic illumination of the operando electrochemical AP-XPS experimental configuration. (B) Co 2p_3/2 XPS spectra recorded under hydrated conditions (18 Torr water vapor) as a function of applied electrochemical potential, showing dynamic changes in oxidation state. Reproduced with permission from ref [66]. Copyright 2017, American Chemical Society. (C) Operando AP-XPS spectra of the Ni 2p_3/2 photoelectron peaks under varied conditions, illustrating the effect of the environment on Ni oxidation states. Reproduced with permission from ref [67]. Copyright 2017, American Chemical Society. (D) Ni K-edge XANES spectra of the Ni/C and Ni₂Fe/C samples at the open-circuit voltage (OCV) state, with the reference spectra of Ni²⁺, Ni³⁺, and Ni⁴⁺ samples. (E) Operando XANES spectra collected at the Ni K-edge for Ni₂Fe₁/C. (F) Oxidation-state distribution for Ni/C and Ni₂Fe₁/C, deconvoluted using a linear combination method. Reproduced with permission from ref [69]. Copyright 2022, Cell Press. (G-H) Operando XANES spectra of Ni − HAB/carbon fiber paper (CFP) under O₂-saturated (G) and Ar-saturated (H) electrolytes during the ORR process, highlighting differences in electronic structure. (I) Coordination numbers extracted from Extended X-ray Absorption Fine Structure (EXAFS) (upper panel) and XANES Ni K-edge positions (bottom panel) for Ni − HAB/CFP under O₂- and Ar-saturated conditions. Reproduced with permission from ref [14]. Copyright 2022, American Chemical Society

Fig. 5

Monitoring the reaction intermediates using operando Raman spectroscopy. (A) SEM images of Fe NiO/NiS₂. (B-C) Potential-dependent operando Raman spectra of Fe NiO/NiS₂ (B) and NiO/NiS (C). Reproduced with permission from ref [75]. Copyright 2022, Wiley. (D) TEM image of Mn₃O₄ NPs with the size of 4 nm. (E) Potential-dependent operando Raman spectra with increasing applied potential in 1 M KHCO₃ electrolyte. (F) Analysis of Raman peak positions. Reproduced with permission from ref [16]. Copyright 2021, Wiley

Fig. 6

Monitoring the reaction intermediates using operando SERS spectroscopy. (A) TEM image of Pt₃Co nanoparticles and Pt₃Co-on-SHINs satellite nanocomposites. (B) Operando electrochemical-SHINERS (EC-SHINERS) spectra of dealloyed Pt₃Co nanocatalysts in 0.1 M HClO₄ with H₂O and D₂O solution saturated O₂. (C) Normalized Raman intensities of Pt − O (black square) stretching mode, and *OOH (red sphere) at different potentials. Reproduced with permission from ref [81]. Copyright 2019, Wiley. (D) Raman spectra of 55 nm Au@2.5 nm Ru surface under various alkaline HER potentials. (E) Raman spectra of 55 nm Au@2.5 nm Ru surface at -0.35 V in different electrolytes. (F) Normalized Raman intensities (blue) and frequency shifts (red) of the Ru-H band at low-valence state Ru(0) (circle) and high-valence state Ru (between + 2 and + 4) (square) surfaces as a function of potential. Reproduced with permission from ref [82]. Copyright 2023, Springer Nature

Fig. 7

Investigation of intermediates through the IR spectroscopy and online electrochemical mass spectrometer. (A) Typical TEM image and corresponding energy-dispersive X-ray (EDX) mapping of HIFs Pt@PtCu₃. (B-C) In-situ SRIR spectra for *OH (B), and *O as well as *OOH (C). (D) Intensity differences of ORR intermediates at HIFs Pt@PtCu₃ and Pt/C. Reproduced with permission from ref [18]. Copyright 2024, American Chemical Society. (E) High-resolution TEM image of CdS-CNTs. (F) Operando DEMS of H₂ and H₂S during CO₂RR. Reproduced with permission from ref [84]. Copyright 2019, Elsevier

Fig. 8

Investigation of product formation by online EC mass spectroscopy. (A) Chronopotentiometric measurements and ion currents for CO₂ in Vulcan and Ni_xB/C-10 catalysts at various current densities. Reproduced with permission from ref [86]. Copyright 2020, Wiley. (B) Faraday efficiency for CO₂ evolution (FE-CO₂), O₂ evolution (FE-O₂), and electrode potential as a function of time at a constant current. (C-D) Total electrochemical and transferred charges for O₂ and CO₂ evolution of Ru/G-450Red (C) and RuO₂/G-450O_x (D), respectively. Reproduced with permission from ref [87]. Copyright 2021, American Chemical Society. (E) Online DEMS measurements of Cu, Ag, Cu/30Ag, and the physical mixture during the CORR. Reproduced with permission from ref [88]. Copyright 2024, Springer Nature. (F) In-situ DEMS measurements for electrocatalytic NO₃RR at − 0.7 V over four continuous cycles on ZnPc MDE. Reproduced with permission from ref [89]. Copyright 2024, Springer Nature

Fig. 9

Typical workflow of data mining-driven catalyst design

Fig. 10

Examples of compositional and geometric descriptors being key features. (A) The linear relationship of overpotential and geometric descriptor μ/t for ABO₃ oxide perovskites, where μ and t are the octahedral and tolerance factors, respectively. Inset is the schematic structure of the ABO₃ perovskite, where the A, B, and O atoms are in black, red, and blue, respectively. (B) The linear sweep voltammetry (LSV) curves of four recommended oxide perovskites, including Cs_0.4La_0.6Mn_0.25Co_0.75O₃, Cs_0.3La_0.7NiO₃, SrNi_0.75Co_0.25O₃, and Sr_0.25Ba_0.75NiO₃. Reproduced with permission from ref [123]. Copyright 2020, Springer Nature. (C) The elemental descriptors for CO absorption sites on the surface of bimetallic electrocatalysts, including elemental atomic number (Z), the Pauling electronegativity (χ), and the coordination number (CN) of the element. (D) The visualization of 131 facets with near-optimal ∆E_CO identified by ML as potential intermetallic catalysts for CO₂ reduction. Reproduced with permission from ref [124]. Copyright 2018, Springer Nature

Fig. 11

Examples of physiochemical descriptors being key features. (A) The classification of 14 descriptors into five descriptor families, including covalency (green), electrostatics (gray), structure (yellow), exchange interaction (red), and electron occupancy (dark gray). (B) The relative importance of 14 descriptors analyzed by penalized regression models. (C) The predicted relative OER activity for ABO₃ perovskites. The overall trend shows that the oxides with Fe, Co, Ni, and Cu exhibited higher OER activities than those with V, Cr, and Mn. Reproduced with permission from ref [128]. Copyright 2015, ACS publication. (D) The structure diagram of single atom doped 2D GaPS₄ and the transition metal dopant candidates for S₁ and S₂ sites. (E) The feature importance of single atom doped 2D GaPS₄ materials for HER. Reproduced with permission from ref [129]. Copyright 2023, ELSEVIER publication

Fig. 12

Screening active electrocatalysts using the ML framework. (A) Stepwise generation of absorption energy database of *O, *OH, and *OOH on 3d, 4d, and 5d TM subnano clusters. (B) Selected descriptors, including d-band specific, geometric, elemental, and electronic features. (C) Model training, evaluation, and interpretability. (D) DFT validation for predicted electrocatalysts. Reproduced with permission from ref [131]. Copyright 2024, American Chemical Society

Fig. 13

Analysis of cardinal values of the Tafel slope with the Bayesian algorithm. (A) The relationship between reported and mean a posteriori (MAP) estimated Tafel slopes. (B) Cumulative distribution function of reported Tafel slope (blue) and their refitted result (red). (C) The effect of physical nonidealities on kernel density estimates (KDE) of the probability distributions (PDF) over the Tafel slope. Reproduced with permission from ref [135]. Copyright 2021, Springer Nature

Fig. 14

Domain knowledge from first-principles based calculations for interpretable models. (A) Schematic representation of chemical bonding on TM surfaces in d-band theory. (B) The plot of DFT-calculated adsorption energies as a function of model-predicted adsorption energies of *OH at the {111}-terminated intermetallic atop sites. Reproduced with permission from ref [137]. Copyright 2020, Springer Nature. (C) True versus predicted energy of Ni-Co-Fe-Pd-Pt. (D) Relationship between ΔG_OH and ΔG_OOH on NiCoFePdPt and RuRhPdIrPt. Reproduced with permission from ref [138]. Copyright 2023, Elsevier

Fig. 15

Integrating domain knowledge and explainable artificial intelligence for interpretable models. (A) Expanded Grad-CAM attention map for Ag electrocatalysts. (B-C) Tafel plots and dLSV plots of randomly selected LSV curves. Specific transition points between kinetic and diffusion-limited regions are represented as a black line on dLSV plots. (D-F) Expanded Grad-CAM attention map, Tafel plots, and dLSV plots for Ni – N/C. Reproduced with permission from ref [139]. Copyright 2025, American Chemical Society

Fig. 16

Integration of in-situ experiments with data mining. (A) Operando Ni K-edge XANES spectra of the heat-treated Ni TMNC catalyst collected before and after the CO₂RR. (B-C) Extracted XANES spectra (B) and corresponding concentration profiles (C). (D) The proposed reaction mechanism based on the PCA analysis and EXAFS fitting. Reproduced with permission, from ref [143]. Copyright 2023, American Chemical Society. (E) Operando Co K-edge XANES for the Co_2.25Fe_0.75O₄ catalyst under OER conditions. (F) Evolution of the weights of the PC-1, PC-2, and PC-3 for the Co_2.25Fe_0.75O₄ under operando condition. (G) Evolution of RDFs for tetrahedrally and octahedrally coordinated Co sites in Co_2.25Fe_0.75O₄ under operando condition. Reproduced with permission, Reproduced with permission, from ref [141]. Copyright 2023, American Chemical Society