Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain
the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in
Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles
and JavaScript.
Huan Guo and coworkers explore the relationship between symmetry, work and efficiency for macroscopic thermodynamic cycles. The results provide insights to design more efficient energy conversion cycles by enhancing the cycle symmetry.
Analysis of the aircraft structural load needs costly and inefficient ground tests. Chenxi and coworkers report a deep learning based approach to predict aircraft strains and loads by identifying the key flight parameters in the load prediction, providing a more efficient and economical way for aircraft load monitoring.
Malcolm White and colleagues apply a lightweight interpretable machine learning framework, FastMapSVM to the classification of seismograms. The authors demonstrate that the approach outperforms neural network alternatives when train data or time is limited.
Medrano and coworkers quantitatively assess the economic value added by lower-limb exoskeletons and their assistance by allowing users to bid for their continued use in an auction scenario. This interdisciplinary investigation offers insights in the impact of exoskeletons to optimise user experience.
Alessandro F. Rotta Loria explores the impact of subsurface urban heat islands on ground deformations as well as civil infrastructure performance. The results provide a meaningful reference to take account of rising ground temperatures for future civil construction design.
Kai Junge and colleagues designed a soft sensorized physical twin of a raspberry plant which they use to collect force data on fruit picking to train a robotic harvester. Early field demonstrations showed promise in rapid training of a robot for the delicate task of soft fruit picking.
Kristen Brosamer and colleagues demonstrate a sensitive lateral flow assay platform using the chemistry of glowsticks to report the detection of human chorionic gonadotropin and SARS-CoV-2 nucleoprotein, read by an unmodified smartphone. Glow LFA is also shown to be a promising platform for the multiplexed detection of analytes on a single test line.
Rachith Kumar and coworkers report a bio-inspired coating technique able to deposit uniform films with thicknesses below 15 nm on various substrates. This method will not require the use of extended ink formulation or high substrate temperature as existing techniques do, potentially reducing the fabrication cost of future electronic devices and batteries.
Devendra Pal and coworkers report an imaging system using Nano-Digital in-line Holographic Microscopy (NanoDIHM) to detect airborne viruses in droplets and aerosols in real time. This system is able to detect various viruses in air, water and heterogeneous matrices within one minute, enabling real-time tracking of pollutant particles for efficient epidemic management.
Perching allows aerial vehicles to maintain high vantage points for prolonged periods with less power consumption. Yi-Hsuan Hsiao and colleagues combine airflow-surface interactions with a mussel-inspired wet adhesive in a lightweight perching design strategy for small drones to improve flight endurance and energy consumption efficiency.
Tom Dillon and colleagues introduce ‘FenFit’, a new computational program designed for automatically fitting fenestrations onto commercially available endografts. This innovation offers promising opportunities to enhance clinical outcomes by providing a user-friendly interface that is quick and significantly less prone to human error.
Surgical application of force for example in incision is a central surgical skill, yet its practical assessment remains largely subjective. Artūras Straižys and colleagues develop a generative model of excision force along with a sensorised scalpel that enables data collection and analysis of manipulation skill during a surgical procedure.
A water harvesting strategy utilizing a hygroscopic lithium chloride impregnated cellulose scaffold yields high water harvesting rate with low energy input over a wide range of relative humidity. The strategy, reported by Wenkai Zhu, Yun Zhang, Chi Zhang and coworkers, provides a potential solution to the global water scarcity problem particularly in arid areas.
Ziqing Lu and colleagues employ deep learning to establish real-time predictions of three-dimensional temperature fields at the millimetre scale during continuous casting, a process widely used in metallurgical manufacturing. The researchers go on to combine the model with Bayesian optimisation for intelligent adaptation of operating parameters to fulfil various industrial production demands.
5G network operators need data traffic predictions to plan network expansion schemes. Yuguang Yang and colleagues demonstrate performance improvement over state-of-the-art forecasting tools of a deep learning model, Diviner. They demonstrate detailed months-level forecasting for massive ports with complex flow patterns.
Karapiperis and Kochmann develop a graph neural network-based model to predict the propagation of cracks in disordered cellular materials. The approach is then applied in the design of architected structures with optimized fracture properties.
Oxygen evolution electrocatalysts for proton exchange membrane water electrolysis encounter degradation even at moderate cell potentials. Piñeiro-García and colleagues develop a quaternary Sn-Sb-Mo-W mixed oxide scaffold to protect RuO2 against early dissolution under harsh acid conditions, extending the lifetime of catalysts as well as the titanium supports used in water electrolysis cells.
Jiaqi Lu et al. apply an asynchronous-parallel recurrent neural network to predict the yield of separating copper and poly(vinyl chloride) components from cable waste by ball milling. The modelling approach could guide the process design to maximise material recovery.
The increasing complexity of the implementation and operation of deep learning techniques hinders their reproducibility and deployment at scale, especially in healthcare. Pati and colleagues introduce a deep learning framework to analyse healthcare data without requiring extensive computational experience, facilitating the integration of artificial intelligence in clinical workflows.
