Causal random forests provide efficient estimates of heterogeneous treatment effects. However, forest algorithms are also well-known for their black-box nature, and therefore, do not characterize how input variables are involved in treatment effect heterogeneity, which is a strong practical limitation. In this article, we develop a new importance variable algorithm for causal forests, to quantify the impact of each input on the heterogeneity of treatment effects. The proposed approach is inspired from the drop and relearn principle, widely used for regression problems. Importantly, we show how to handle the forest retrain without a confounding variable. If the confounder is not involved in the treatment effect heterogeneity, the local centering step enforces consistency of the importance measure. Otherwise, when a confounder also impacts heterogeneity, we introduce a corrective term in the retrained causal forest to recover consistency. Additionally, experiments on simulated, semi-synthetic, and real data show the good performance of our importance measure, which outperforms competitors on several test cases. Experiments also show that our approach can be efficiently extended to groups of variables, providing key insights in practice.
Scale-resolving fluid dynamics simulations require fine grids and high-order numerical schemes. For these applications, explicit time integration is usually employed, which severely limits the time step size. Although implicit time integration schemes can have favorable stability properties and allow for larger time steps, they are often overlooked due to their higher computational cost. Implicit schemes are instead commonly used in steady-state simulations, where low-order discretizations are preferred. This work explores the use of state-of-the-art high-order implicit temporal schemes with high-order spatial methods on unstructured grids. Advection and diffusion test cases demonstrate notable mesh convergence rates even with large time steps. An advection-diffusion-reaction numerical benchmark, representative of a spherical flame, is introduced and shows improved accuracy with advanced implicit schemes. Diagonally implicit Runge-Kutta methods, in particular, offer a significant gain in cost-accuracy trade-off over explicit methods, especially in diffusion-driven scale-resolving computations.
Blind deconvolution tackles the issue of recovering a signal from a convolution between an initial signal and a filter with an unknown kernel. To address the ill-posed nature of blind deconvolution, we leverage the sparse characteristics of the signals in a pre-existent dictionary. Rather than imposing sparsity directly on the signal using L0 or L1 penalties, we express it as a prior on the signal’s covariance matrix. The hierarchical prior acts like a decoupling between the signal and its sparsity, making estimation a classical a posteriori problem. The proposition revolves around a maximum a posteriori estimation in an Expectation - Maximization framework for alternate optimization of the signal and the filter. We give simulation results in comparison with MAP oracle values for any sparse basis.
High-temperature proton exchange membrane fuel cells (HT-PEMFCs) are interesting alternatives to fossil fuel-based technologies for power generation. Their higher operating temperature compared to classic low temperature (LT) PEMFCs is highly beneficial regarding the total system complexity and weight, especially for applications which cannot involve bulky/heavy cooling systems, like aeronautics. However, at the current state-of-art of the technology, the membrane electrode assembly performance in HT-PEMFC does not reach that of LT-PEMFCs and needs to be improved; this requires better catalyst material, catalyst layer structure and membrane doping [1], [2]. A given catalyst performance is largely influenced by phosphoric acid electrolyte poisoning [3], [4] and some studies have already evaluated such poisoning level on various catalysts toward oxygen reduction reaction [5], [6]. The current study further addresses this issue and specifically aims to understand the impact of various parameters which should impact the catalyst activity in H 3 PO 4 electrolytes (loading, nano-particle size and shape, chemistry of the catalyst particles (Pt vs PtNi), density of aggregates, nature of the carbon support) and to evaluate the poisoning effect whether at low potential (anode) or at high potential (cathode). The effect of different parameters was unveiled thanks to a catalyst library (Figure a), analyzed comparatively in 1 M HClO 4 and 1 M H 3 PO 4 electrolytes at room temperature with a classic rotating disk electrode set-up and a gas diffusion electrode set-up. Pseudo CO-Stripping voltammetry, Hupd and CO-stripping voltammetry enable to shed light on the poisoning of the Pt surfaces: the electrochemical surface area (ECSA) is divided by a factor around 2 in H 3 PO 4 electrolyte compared to HClO 4 and varies according to the catalyst properties. The impact on hydrogen oxidation reaction is then noticeable. The poisoning at high potential (> 0.6 V vsRHE ) was mainly evaluated thanks to the ORR activity (Figure b): the Pt nanoparticle size/shape, their loading on the carbon support and alloying with Ni do impact their ORR activity. All the catalysts exhibit significantly lower activity in H 3 PO 4 than in HClO 4 (by a factor ca 10). The mechanisms of H 3 PO 4 -induced poisoning and potential strategies to mitigate it will be detailed. [1] S. S. Araya et al. , “A comprehensive review of PBI-based high temperature PEM fuel cells,” Int J Hydrogen Energy , vol. 41, no. 46, pp. 21310–21344, Dec. 2016, doi: 10.1016/j.ijhydene.2016.09.024. [2] R. E. Rosli et al. , “A review of high-temperature proton exchange membrane fuel cell (HT-PEMFC) system,” Int J Hydrogen Energy , vol. 42, no. 14, pp. 9293–9314, Apr. 2017, doi: 10.1016/j.ijhydene.2016.06.211. [3] B. F. Gomes et al. , “Effect of phosphoric acid purity on the electrochemically active surface area of Pt-based electrodes,” Journal of Electroanalytical Chemistry , vol. 918, Aug. 2022, doi: 10.1016/j.jelechem.2022.116450. [4] N. Sugishima et al ., “Phosphorous Acid Impurities in Phosphoric Acid Fuel Cell Electrolytes: I . Voltammetric Study of Impurity Formation,” J Electrochem Soc , vol. 141, no. 12, pp. 3325–3331, Dec. 1994, doi: 10.1149/1.2059334. [5] Q. He et al , “Influence of phosphate anion adsorption on the kinetics of oxygen electroreduction on low index Pt(hkl) single crystals,” Physical Chemistry Chemical Physics , vol. 12, no. 39, pp. 12544–12555, Oct. 2010, doi: 10.1039/c0cp00433b. [6] K. ‐L. Hsueh et al , “Effects of Phosphoric Acid Concentration on Oxygen Reduction Kinetics at Platinum,” J Electrochem Soc , vol. 131, no. 4, pp. 823–828, Apr. 1984, doi: 10.1149/1.2115707. Figure 1
This study aims at presenting the current status of the IN718 development using Binder Jetting at Safran and the challenges associated with its industrialization for aeronautic applications.Printing tests done on several laboratory/production machines, with powders from different suppliers and with various particle size distributions will be used to exemplify the impact of such parameters on the part quality at each stage, from printing to sintering. A study of part density, sintering, reproducibility has been performed as well. Based on these findings, a map of influent parameters will be shared to elucidate the path towards industrialization and certification of parts in the aeronautic field. Perspectives will be given regarding the common framework to be developed for the Binder Jetting to become the cost efficient and high material quality process.
