Manufacturing time estimation is a knowledge-intensive task in Engineer-To-Order (ETO) industries, where product diversity and specification variability impose a significant cognitive workload on engineers. This is particularly critical for suppliers who need to estimate their offers in order to respond to calls for tenders or direct inquiries from customers. Unlike most existing studies, which apply pure machine learning models, this study presents an integrated framework that involves an experience-based reasoning mechanism as the main component. Additionally, it makes a notable contribution by fully automating the manufacturing time estimation for multifamily ETO products within a single framework. The proposed framework brings two technical contributions: (i) automated product family identification using Rule-Based Learning (RBL) or Artificial Neural Network (ANN) and (ii) improved case retrieval and manufacturing time estimation in Case-Based Reasoning (CBR) by leveraging product family-guided case base structuring and KD-Tree for search space reduction. The framework is validated on real-world manufacturing data from a French metallurgy company, demonstrating performance in estimation accuracy and computational efficiency, highlighting its practical viability and generalizability across ETO manufacturing. It can be easily integrated with existing enterprise information systems, such as ERP/MES, to support the estimation of offers or operations management.
Proton exchange membrane fuel cells play a vital role in achieving a hydrogen-powered, carbon-neutral society. Although they have been deployed in stationary and transportation applications, their limited durability and reliability still restrict large-scale commercialization. Assessing internal health states is critical to overcoming these challenges by preventing operational faults and minimizing degradation rates. However, hydrogen fuel cells are highly nonlinear, multiscale, and multiphysics systems, hindering the estimation of system health states. This review investigates a nondestructive and in situ characterization technique, electrochemical impedance spectroscopy (EIS), to characterize the health states of hydrogen fuel cells. The variation in internal health states is directly reflected in the impedance spectrum, which can be assessed through proper interpretation. Nevertheless, developing EIS for hydrogen fuel cells is a nontrivial task. Implementing and applying EIS requires interdisciplinary studies in signal processing, control theory, and machine learning in particular cases. This review systematically discusses the implementation, interpretation, and application of EIS for hydrogen fuel cells. The onboard applicability and short measurement time are key factors in EIS implementation. Model-based and DRT-based interpretation approaches are reviewed for efficiently extracting valuable health states. Based on these methods, the applications of EIS in performance characterization, fault diagnosis, degradation analysis, and prognostics are discussed. Four perspectives regarding general, non-electrical, localized, and multiphysics-integrated EIS, are proposed to highlight and guide EIS-related studies in hydrogen fuel cells. This review provides a comprehensive overview of the deployment and application of EIS in hydrogen fuel cells and offers insights into addressing durability and reliability challenges.
Systemic lupus erythematosus (SLE) exhibits marked clinical and biological heterogeneity that is poorly captured by group-based analyses. Using the MRL/Lpr model, we describe an individual mouse displaying a severe phenotype, and provide a characterization combining clinical assessment, inflammatory profiling, neuroaxonal injury, and targeted metabolomics. Despite severe disease, this animal did not display a globally exaggerated cytokine profile relative to other MRL/Lpr mice, although TNFα and IL-1β ranked among the highest in the cohort. In contrast, plasma neurofilament light chain (NfL) was markedly elevated, and metabolomic profiling revealed pronounced disruptions in urea-cycle intermediates, sulfur amino-acid metabolism and muscle-derived metabolites, together with a broad depletion of circulating amino acids. These observations raise the possibility that, at advanced disease stages, circulating cytokine levels may plateau and lose discriminative capacity, whereas metabolic readouts may retain greater capacity to track disease severity.
Accurate pre-harvest grape yield estimation is a critical challenge in precision viticulture, yet current optical approaches (RGB cameras, LiDAR) are fundamentally limited by foliage occlusion and sensitivity to ambient lighting conditions. This paper presents a ground-based grape yield estimation system using a dual 77 GHz Multiple-Input Multiple-Output (MIMO) Frequency-Modulated Continuous-Wave (FM-CW) radar mounted on a rover operating at an average speed of 1.0 m/s (3.7 km/h), matching practical vineyard tractor speeds. Unlike prior radar-based studies conducted in static or slow-moving setups, the proposed pipeline achieves continuous 3D reconstruction of vinerows through digital beamforming in the elevation plane, combined with an IMU-fused Kalman filter for point cloud stabilization on irregular terrain. An experimental campaign was conducted on 136 km of scanned vinerows across seven grape varieties and three phenological growth stages (BBCH 75–89) in a commercial vineyard in Gascony, France. Statistical features extracted from 3D radar echo level images are used to train a Support Vector Regressor (SVR) evaluated in cross-validation. The primary out-of-sample result is a cross-validation Mean Absolute Error (CV MAE) of 31.2% on the full multi-variety dataset without any defoliation, achieved against certified wine-vat bulk weights as the ground truth reference. This figure should be contextualized against the ∼ 24–25% discrepancy observed between portable field weighing plates and vat-certified weights across the same dataset, and against manual pre-harvest estimation errors routinely exceeding 30% in viticultural practice. A sensitivity analysis over physical parameters reveals that grape variety is the dominant −and limiting- source of performance heterogeneity: CV MAE ranges from 7% for Tannat to 69% for Baco, demonstrating that a single global model is insufficient for multi-variety deployment and that variety-specific calibration is a necessary condition for operational use. These results are best understood as an architectural proof of concept: the primary contribution is the demonstration that MIMO digital beamforming removes the speed bottleneck of prior mechanically-scanned radar systems, enabling gapless 3D vinerow reconstruction at practical field speeds. The yield estimation results provide a first quantitative characterization of the system’s sensing capability and its variety-dependent limitations and establish a baseline for future variety-specific model development.
Ab initio thermodynamics of water on flat Ru(0001) predicts full dissociation to atomic O and H, with no stable hydroxyl, contradicting the experimental consensus that OH is an essential intermediate on Ru/TiO2 hydrodeoxygenation catalysts. Grand-canonical sampling of a sub-nanometer Ru10/TiO2(101) cluster across 0 ≤ nH2O ≤ 9 on pristine and oxygen-vacancy supports shows hydroxyl returning as the dominant adsorbate-O species (45–50%), bound at single-Ru top sites on under-coordinated periphery atoms with no flat-metal analogue. An OH-rich 6-water phase is the predicted operational state across the 400–530 K aqueous-catalysis window. Hydrogen spillover to the support is an order of magnitude more probable on the reduced support at saturation coverage, but lies below the sampling resolution at the operational six-water plateau on both supports; Bader analysis shows the vacancy also changes its character, with the transferred hydrogen arriving hydridic on an under-coordinated Ti rather than protic on a lattice oxygen. The active site is the cluster periphery and the relevant adsorbate is OH, not the fully dissociated O + H of the flat-Ru paradigm.
