20005 research outputs found
Sort by
RTK-over-LoRaWAN: high-precision positioning service for low-power IoT devices
No full textCentimeter-level positioning accuracy provided by real-time kinematic (RTK) systems can support emerging IoT applications that require high-resolution spatial monitoring and precise geo-referenced sensor reporting. However, the high data rate requirements of traditional RTK protocols prevent their use in IoT networks with limited communication capacity and regulatory transmission constraints, such as LoRaWAN (Long Range Wide Area Network). We propose an RTK-over-LoRaWAN architecture that significantly reduces the data volume required for RTK positioning through adaptive RTCM scheduling. We conduct extensive field experiments with baseline distances up to 10km to validate that our method maintains centimeter-level accuracy under sparse RTCM delivery. The results show that RTK initialization requires only 500 B to 1000 B of total downlink payload, while extended RTK tracking is sustained with an effective downlink throughput below 15 B/s, reducing data requirements to under 5% compared to conventional RTK implementations. Our approach meets regulatory constraints such as those defined for the EU868 region, achieving duty cycles well below 1% per hour and enabling the practical deployment of RTK-based positioning in LoRaWAN networks
Development of a low cost AI-capable drone for obstacle avoidance: a tutorial
No full textAs part of our research, we present a concept for an AI-enabled multi-rotor research aircraft for university research purposes. With a focus on cost efficiency without sacrificing performance, we have developed and tested the AI-enabled drone for monocular depth estimation (MTS) and obstacle avoidance in an in-door scenario under controlled test conditions (lighting conditions, take-off position and homogeneity of obstacles). A comprehensive GPS or IPS system is not required. This tutorial covers the essential aspects of developing and building camera drones, focusing on the integration of deep learning algorithms, sensor technologies and robust flight control systems. The MTS is limited to an in-door environment to prepare students for the research field of perception and provide opportunities for further development. An intuitive user interface in Python was developed as a basis. The proposed configuration is based for the first time on bidirectional communication between the Flight Control Computer (FCC) of the Beaglebone Blues (BBB) and the Flight Companion Computer (FCC) NVIDIA Orin Nano via a UART interface. The presented user interface was developed inhouse and is capable of adjusting the controller's parameters on-the-fly via a WiFi connection. This tutorial is intended to provide researchers, developers and students interested in developing flying robots for obstacle avoidance with insights into costeffective design strategies and the integration of AI technologies to improve the capabilities of drones
Securing blockchain-based distributed learning for heterogeneous clients through knowledge distillation and zero-knowledge proofs
No full textDistributed learning (DL) is gaining popularity as it enables clients (e.g., AI Agents) to enhance their machine learning (ML) models’ performance by exchanging knowledge without revealing private datasets. State-of-the-art DL approaches primarily focus on transferring knowledge between heterogeneous clients with diverse model architectures, connecting clients with those that can improve their models, and protecting data privacy. However, they overlook the threat of malicious clients that potentially downgrade the models’ performance by sharing inaccurate knowledge or excluding high-performing clients from the training procedure.Therefore, we introduce the Zero-Knowledge Blockchain-Based Knowledge Distillation Learning Framework (zkBKD). In zkBKD, heterogeneous clients communicate with a blockchain network to discover high-performing clients, verify zero-knowledge proofs (ZKPs) to ensure the correctness of the knowledge shared from other clients, and vote to eliminate malicious clients. We analyze security and privacy risks and show that zkBKD prevents membership, poisoning, and collusion attacks. We conduct extensive experiments on two standard datasets across heterogeneous clients with four model architectures. The experimental results demonstrate that zkBKD relatively improves the average model accuracy of all clients by 25.71%. Even lightweight models such as ResNet-2 achieve up to a 103.35% accuracy gain compared to independent training
Rainfed spring canola yield response to changing heat and water stress in the Canadian Prairie region
No full textCanola is a significant crop in Canadian agriculture and the economy. However, Canada’s average temperatures have risen rapidly over the past eight decades, changing temperature patterns and water availability for canola production. This study aims to explore the impacts of air temperature and soil water availability on spring canola production from 2025 to 2050. Accordingly, this study introduces DSSAT calibration and simulation of the current hybrid InVigor®L340PC, integrating the Shared Socioeconomic Pathways. Leveraging DSSAT-Pythia, gridded simulations capture spatial variability in water and temperature stress interactions, driven by a large ensemble of climate models. The analysis reveals how precipitation and temperature changes jointly influence spring canola development. Yield projections under these conditions provide critical insights into the future viability of rainfed spring canola and inform adaptation strategies for growers and policymakers. Findings demonstrate negative impacts on exclusively rainfed spring canola production in the Canadian Prairie Region under diverse climate scenarios from 2025 to 2050. The main canola growing ecozone (Aspen Parkland) is expected to have higher air temperatures and lower soil water content if greenhouse gas emissions keep rising. An average increase of 1.5◦C in air temperature and 0.025 in the water stress factor indices may result in annual yield reductions of 203 ± 4.3 and 121 ± 13.6 kg ha⁻¹, in Lake Manitoba Plain and Aspen Parkland ecoregions, respectively. Given that future canola production is expected to continue in the same ecoregions it is recommended that adaptation and mitigation strategies are developed and adopted to improve canola production conditions in these ecoregions
CapAware: capacity-aware uplink bandwidth prediction for cellular networks
No full textAs remotely controlled and autonomous vehicles become widely available, the demand for high Quality of Service over cellular networks for their remote control and monitoring is becoming increasingly important. Accurate prediction of available uplink bandwidth is essential to mitigate bandwidth fluctuations and avoid impacting real-time applications, ensuring reliable and low-latency video streams. In particular, bandwidth overpredictions lead to packet losses, retransmissions, and significant latency increases, especially during network handovers, as network buffers fill up. Prior bandwidth prediction approaches lower absolute or relative errors but fail to address the impacts of overpredictions and the associated latency spikes.This paper introduces CapAware, a bandwidth prediction approach explicitly designed to minimize capacity violations (i.e., overpredictions) and reduce latency spikes during network handovers for uplink streams. It utilizes an efficient neural network architecture with an integrated handover prediction mechanism and a learnable capacity-aware loss function. CapAware predicts network handovers with a 92.4% F1 score and improves efficiency by 24.4% using its custom loss function with predicted handover information. Compared to deep-learning baselines, CapAware improves network efficiency (i.e., utilizationto-capacity violation ratio) by 4.7% and 34.9% on 5G SA datasets
Künstliche Intelligenz in der Produktion
No full textDer Beitrag zielt darauf ab, einen Einblick in die technologische Dimension von Künstlicher Intelligenz zu geben. Konkret werden potenzielle KI-Technologien, KI-Lernansätze und KI-Anwendungsfälle für den industriellen Kontext vorgestellt. Bei den KI-Technologien werden regelbasierte KI – auch als symbolische KI bezeichnet, Machine Learning (ML), Neuronale Netze (Deep Learning), Natürliche Sprachverar-beitung (NLP) sowie Computer Vision näher vorgestellt. Bei den Lernansätzen werden drei klassische Lernansätze des überwachten Lernens, des unüberwachten Lernens sowie des bestärkenden Lernens beschrieben. Eine Darstellung möglicher Use Cases im industriellen Kontext, bei der auch die Chancen und Risiken dieser KI-Systeme dargestellt werden, schließen den Beitrag ab
Comparative study of phytoplasma DNA extraction methods
No full textThe study of phytoplasmas and methods of their detection is an important aspect of determining the phytosanitary condition of the crops in a territory and to help the use of uninfected planting material by agricultural producers. For the study of phytoplasmas it is necessary to carefully and correctly prepare the sample to be analysed with appropriate DNA extraction procedures. Sample preparation consists in extracting fragments of tissues from suitable plant material mainly consisting of phloematic tissues. A comparative study and validation of methods of DNA extraction of phytoplasmas on the example of vegetative parts of the pear tree was carried out. It is shown that the reagent kits CytoSorb and SORB-GMO-B minimize the testing time and are not inferior to the traditional method based on CTAB extraction
An integrated design process for lightweight AM products using the DED process
No full textOne challenge facing aviation today is the reduction of greenhouse gases. By reducing the weight of the aircraft, fuel consumption and the associated emissions can be reduced. For aircraft cabin interiors so-called sandwich structures are established. A major issue with these sandwich constructions is the load introduction, often causing over dimensioning, which counteracts the desired lightweight goals. Additive manufacturing offers new possibilities for structural design and customisation due to its design freedom. By using the Direct Energy Deposition process, which only requires a local inert gas atmosphere, components with large dimensions are feasible. However, this manufacturing process results in great challenges regarding accuracy, manufacturability, e.g. the use of closed cross-sectional profiles and the positioning of the build platform. In this paper a process to tackle these issues is developed. Applying the developed process, a design of a cabin partition is developed. From this design, bending specimens are derived, manufactured and tested to evaluate the feasibility and validate simulation models. It emerges that circular cross-sections in particular should be preferred for the next cabin partition design
The role of wind velocity in saline water evaporation from porous media and surface salt crystallization dynamics
No full textSaline water evaporation from porous media with the corresponding surface salt crystallization patterns play a vital role in many environmental and engineering applications. While the impact of factors such as type and concentration of salt, particle size and angularity, and ambient temperature and humidity are relatively well characterized [1]–[3], the influence of wind and aerodynamic conditions on saline water evaporation and salt crystallization is not fully understood. We conducted a series of laboratory experiments in a wind tunnel to systematically investigate the effect of wind flow on saline water evaporation and dynamics of salt crystallization. Cylindrical sand columns (D: 5 cm – H: 20 cm) were placed in the test section of the wind tunnel. Surface of the samples were exposed to uniform mean wind velocities of 0.5 and 5 m/s. To keep samples fully saturated during the evaporation experiments, sand columns were supplied from Mariotte bottles containing 10, 15, and 20% NaCl solutions. Evaporation rates were monitored by measuring mass losses from Mariotte bottles, while salt crystallization patterns were captured using an optical camera positioned above the surface of columns. Preliminary results indicate that variation in aerodynamic conditions and turbulence patterns, driven by changes in wind velocity and surface roughness (due to crystal growth), significantly alter evaporation rates and salt crystallization process. Distinct crystallization patterns were observed with variation of wind velocity with possible influences on the evaporative fluxes. Using the measured data, we will identify the key effects of air flow regimes coupled with the salt concentration on evaporative losses and the evolution of crystallized salts at the surface, which will be important for a wide range of environmental and hydrological applications.Deutsche Forschungsgemeinschaft (DFG