Johnson Matthey (United Kingdom)

Industrial research and development (R&D) of advanced materials, particularly catalysts for sustainable technologies, is fundamentally hindered by a data analysis bottleneck. State-of-the-art characterisation instruments generate vast and complex datasets, the processing of which has become a significant impediment to the pace of innovation. This paper introduces the Johnson Matthey Data Science App Store, a strategic digital platform developed to overcome this challenge. The platform is a centralised, web-based suite of data science applications designed to democratise advanced analytics through intuitive, user-friendly graphical interfaces. The usefulness of the approach in generating value is illustrated with examples in three-dimensional device and micron level characterisation, particle size analysis and high-precision strain analysis.

This work analyses the sensitivity of Discrete Element Method (DEM) simulations to contact model parameter values, considering both particle–particle and particle-geometry interactions, and the impact on system response for stirred mills. Determining the correct pairwise parameters in a DEM model remains a non-trivial undertaking. This difficulty is in part due to complex relationships between the parameters and system response; often the parameters themselves are correlated. Differing process regimes may also draw different levels of influence from the contact parameters, e.g. the coefficient of restitution may be deemed insignificant in friction-driven flows, but significant for other regimes. In this study, process settings (mill speed and mass fill) as well as contact model parameters are varied to infer the change in system response. A sensitivity analysis at three levels (empirical, component analysis on simplified derivatives, and considering full derivatives) is presented. It is found that the response of a fully packed system showed little variance with contact parameter values, especially when translational motion is the focus. Furthermore, particle-geometry terms have a greater impact than the particle–particle equivalents. Translational motion is dominated by the coefficient of restitution and rotational motion by the rolling friction coefficient. The corollary is that any parameter estimation should account for the anticipated process conditions, and thus a DEM study should consider sensitivity analysis as a key step of the parameter estimation process. This ensures that variance in contact models is better recognised by estimation methods, and thus provides better parameter estimation. • Densely packed mills show dependency on sliding friction. • Less packed mills show strong dependency on the coefficient of restitution. • Sensitivities vary across particle–particle and particle-geometry contacts. • The contact parameter dependency varies across different mill geometries.

Flows in square channels are common in applications, such as automotive after-treatment systems and heat exchangers. Flows with axial flow entry are well understood, but for oblique flow entry, there is no clarity on the additional pressure loss magnitude or the flow regime. Laminar flow is often assumed, even though flow separation at the channel entrance can cause a transition to turbulence. Here, the impact of oblique flow entry on the flow is investigated using LES (large eddy simulation) and RANS (Reynolds averaged Navier–Stokes) models, and their advantages and limitations are identified. The LES simulations show that the shear layer at the channel entrance produces continuous shedding of eddies that persist downstream even at moderate channel Reynolds numbers (≈ 2000). The LES predicted pressure losses mostly agree with experimental data, and the differences observed are attributed to the difficulty of accurately replicating the experimental geometry. It is shown that both LES and RANS results are sensitive to the rounding of the leading edge (present in experiments). Including edge rounding improves the pressure predictions. RANS simulations mostly agree with experimens, but unlike LES did not predict transitional flow phenomena for sharp leading edge. This study provides insight into the flow structure and sources of pressure losses in square channels, highlights the importance of understanding key flow and geometric features when using LES to predict complex flows involving flow separation and shear layers, and indicates the need to further investigate the complex instability arising from interactions between secondary flow and shear-layer roll-up.

Ammonia is an attractive hydrogen carrier, yet its practical use is limited by the need for efficient catalytic decomposition. We demonstrate that in-situ N-doping of Ru nanoparticles and graphitized carbon nanofiber supports during reaction produces a sharp increase in hydrogen production during the first 40h, followed by stable activity. Spectroscopic and microscopic analyses, together with density functional theory simulations, reveal that Ru nitridation is rapid and support-independent, resulting in a mechanistic shift from the traditional Langmuir-Hinshelwood to a Mars-van Krevelen pathway, further confirmed by isotopic labelling experiments. In contrast, the progressive nitridation of the carbon support, observed via X-ray photoelectron spectroscopy, modulates the electronic environment of Ru and functions as a dynamic nitrogen reservoir that enables reversible N atoms exchange with the Ru particles, facilitating N desorption from the Ru surface and thereby governing the catalytic activity enhancement. These new findings provide new mechanistic insight into ammonia decomposition and establish progressive nitrogen doping of carbon supports as a strategy for designing efficient metal-based catalysts for hydrogen production.

PFSA ionomer thin films were prepared from realistic ionomer solution concentrations at variable thicknesses on SiO2, Pt, and C. The thin films, prepared from two different ionomers, were then characterised using Neutron Reflectometry and Dynamic Vapour Sorption.  All ionomer films exhibited the expected increase in thickness due to hydration on exposure to a high relative humidity. Use of a custom-built environmental control chamber enabled measurement of the fast hydration equilibration times of the films, unlike previous studies that used salt solutions for humidity control. The data obtained enabled the probing of the effect of ionomer film thicknesses, substrate, and ionomer type. The measurements show that the different ionomers, despite taking up similar amounts of water per sulfonic acid group, showed significant differences in film swelling in response to that water uptake. The effect of ionomer type is shown to be greater than differences due to substrate. This study therefore provides mechanistic insight into the differences performance of the catalyst layers containing these ionomers at different ambient humidities beyond traditional characterisation techniques.

Much of the efficiency of PEMFCs is determined by the water handling properties of the cathode-side catalyst layer (CCL), which depend upon its particular pore structure and distribution of the various phases therein. A new (to this application) gas adsorption-based method has been used to assess the global spatial variation in coverage of the catalyst support with ionomer for differing ionomer contents. The rarely-used nitrogen over-condensation method was proven essential for the comprehensive characterisation of CCLs with pores that span a much wider size range than can be seen with conventional adsorption, and is also more statistically-representative than imaging methods. Percolation theory-based models have been used to predict the impact of different saturation levels of the invasion percolating phase (using mercury, as a model for condensed water) on gas phase mass transport rates. A novel scaling analysis has shown that the saturation level and disposition of the invading phase impacts void space accessibility via modifying the apparent diffusional penetration depth for mass transfer. This showed the key importance of surface clusters of larger pores to both water handling and gas phase mass transport. • Method to assess spatial variation in coverage of the catalyst with ionomer. • Gas overcondensation detected macroporous voids. • Percolation model predicted impact of non-wetting fluid saturation on gas transfer.

The Oxygen Reduction Reaction (ORR) is a key reaction in fuel cells and metal-air batteries, where high overpotentials remain a critical challenge despite extensive research. While experimental studies have revealed the importance of surface oxidation, a unified computational framework capable of simultaneously capturing both the thermodynamic aspects of rate-determining steps and the kinetic effects of site-blocking on the overpotential has remained elusive. In this work, we present a novel computational approach that bridges this gap by combining grand-canonical Monte Carlo simulations with the MACE-MP-0 foundation model to study the ORR on experimentally reconstructed Pt nanoparticles. This framework enables the systematic investigation of oxidation effects across multiple scales, from atomic-level place-exchange mechanisms to macroscopic kinetic behavior. Our simulations reveal a strong dependence of system thermodynamics on oxygen coverage and successfully predict the place-exchange mechanism onset at 1.06 V vs. SHE, in agreement with experimental observations. Through established scaling relations and deletion energy analysis, we quantify both the rate-determining step and the distribution of reactive sites on the oxidized surface, providing insight into the complex interplay between surface oxidation and ORR activity. By linking our results with both theoretical and experimental benchmarks on multiple points, we ensure the viability of our assumptions and approach. Using a simplified kinetic model derived from our simulations, we demonstrate agreement with core experimental observations, validating the predictive power of foundation models in computational electrochemistry. This work not only provides a comprehensive understanding of oxide effects in ORR but also establishes a versatile computational methodology that can be readily extended to study similar electrochemical processes on other catalytic systems, offering a powerful new tool for rational catalyst design.

Silica-supported Pd nanoclusters (Pdn/SiO2) are essential catalysts in methane oxidation for pollution abatement. These catalysts operate at high temperature for long periods exposed to the exhaust gases of combustion engines. For catalyst design to ensure the required longevity, insight into the interaction between nanoclusters and the SiO2 surface is essential. Accordingly, there is a need for well-defined, computationally inexpensive, quantum mechanical models of these structures from which cluster support interactions can be understood. In this work, we have found that a strong adsorption site for a single Pd1 atom on the hydroxylated α-SiO2(101) surface can be formed when Pd adsorption is accompanied by the elimination of water. We show that this becomes a thermodynamically favourable process at the temperatures commonly used for calcination following catalyst synthesis using wet impregnation. Using this initial anchor point Pdn and Pdn(xO) nanoclusters are systematically constructed guided by the Mulliken charges of atoms at each stage. We then investigate the surface interactions of these Pdn and Pdn(xO) nanoclusters by considering the gas phase references of the same clusters. Our calculations show stability varies according to cluster size. Oxidation of the clusters leads to a weakening of cluster-surface interaction.

The data analysis workflow for in situ electron microscopy experiments can require a significant amount of human-intensive and repetitive effort. The generation of Python-based scripts that incorporate simple machine learning algorithms are quite well established in biological sciences but not often utilised in the study of catalytic systems. Such scripted analysis is not only more efficient, but readily reproducible and allows a wide range of quantitative results to be reported, including but not limited to average and total particle size, particle counting and particle size distributions. In this work we utilise these tools to examine the effect of cycling reducing and oxidising atmospheres on copper oxide nanoparticles.