Towards Communication-Efficient and Attack-Resistant Federated Edge Learning for Industrial Internet of Things

Federated Edge Learning (FEL) allows edge nodes to train a global deep learning model collaboratively for edge computing in the Industrial Internet of Things (IIoT), which significantly promotes the development of Industrial 4.0. However, FEL faces two critical challenges: communication overhead and data privacy. FEL suffers from expensive communication overhead when training large-scale multi-node models. Furthermore, due to the vulnerability of FEL to gradient leakage and label-flipping attacks, the training process of the global model is easily compromised by adversaries. To address these challenges, we propose a communication-efficient and privacy-enhanced asynchronous FEL framework for edge computing in IIoT. First, we introduce an asynchronous model update scheme to reduce the computation time that edge nodes wait for global model aggregation. Second, we propose an asynchronous local differential privacy mechanism, which improves communication efficiency and mitigates gradient leakage attacks by adding well-designed noise to the gradients of edge nodes. Third, we design a cloud-side malicious node detection mechanism to detect malicious nodes by testing the local model quality. Such a mechanism can avoid malicious nodes participating in training to mitigate label-flipping attacks. Extensive experimental studies on two real-world datasets demonstrate that the proposed framework can not only improve communication efficiency but also mitigate malicious attacks while its accuracy is comparable to traditional FEL frameworks.

The bulk parameterizations of turbulent air-sea fluxes in NEMO4: the origin of Sea Surface Temperature differences in a global model study

Wind stress and turbulent heat fluxes are the major driving forces which modify the ocean dynamics and thermodynamics. In the NEMO ocean general circulation model, these turbulent air-sea fluxes (TASFs), which are components of the ocean model boundary conditions, can critically impact the simulated ocean characteristics. This paper investigates how the different bulk parametrizations to calculated turbulent air-sea fluxes in the NEMO4 (revision 12957) drives substantial differences in sea surface temperature (SST). Specifically, we study the contribution of different aspects and assumptions of the bulk parametrizations in driving the SST differences in NEMO global model configuration at ¼ degree of horizontal resolution. These include the use of the skin temperature instead of the bulk SST in the computation of turbulent heat flux components, the estimation of wind stress and the estimation of turbulent heat flux components which vary in each parametrization due to the different computation of the bulk transfer coefficients. The analysis of a set of short-term sensitivity experiments, where the only experimental change is related to one of the aspects of the bulk parametrizations, shows that parametrization-related SST differences are primarily sensitive to the wind stress differences across parametrizations and to the implementation of skin temperature in the computation of turbulent heat flux components. Moreover, in order to highlight the role of SST-turbulent heat flux negative feedback at play in ocean simulations, we compare the TASFs differences obtained using NEMO ocean model with the estimations from Brodeau et al. (2017), who compared the different bulk parametrizations using prescribed SST. Our estimations of turbulent heat flux differences between bulk parametrizations is weaker with respect to Brodeau et al. (2017) differences estimations.

Coupled Modeling Approach for Laser Shock Peening of AA2198-T3: From Plasma and Shock Wave Simulation to Residual Stress Prediction

Laser shock peening (LSP) is a surface modification technique to improve the mechanical properties of metals and alloys, where physical phenomena are difficult to investigate, due to short time scales and extreme physical values. In this regard, simulations can significantly contribute to understand the underlying physics. In this paper, a coupled simulation approach for LSP is presented. A global model of laser–matter–plasma interaction is applied to determine the plasma pressure, which is used as surface loading in finite element (FE) simulations in order to predict residual stress (RS) profiles in the target material. The coupled model is applied to the LSP of AA2198-T3 with water confinement, 3×3mm2 square focus and 20 ns laser pulse duration. This investigation considers the variation in laser pulse energy (3 J and 5 J) and different protective coatings (none, aluminum and steel foil). A sensitivity analysis is conducted to evaluate the impact of parameter inaccuracies of the global model on the resulting RS. Adjustment of the global model to different laser pulse energies and coating materials allows us to compute the temporal pressure distributions to predict RS with FE simulations, which are in good agreement with the measurements.

The Effect of Excited Species on the Collisional Energy of Argon Inductively Coupled Plasmas: A Global Model Study

Global modeling of inductively coupled plasma (ICP) reactors is a powerful tool to investigate plasma parameters. In this article, the argon ICP global model is revisited to explore the effect of excited species on collisional energy through the study of different approaches to particle and energy balance equations. The collisional energy loss is much more sensitive to modifications in the balance equations than the electron temperature. According to the simulations, the multistep ionization reduces the collisional energy loss in all investigated reaction sets and the inclusion of heavy species reactions has negligible influence. The plasma parameters obtained, such as total energy loss and electron temperature, were compared with experimental results from the literature. The simulated cases that have more excited species and reactions in the energy balance are in better agreement with the experimental measurements.

A Global Debris Flow Susceptibility Map Based on a Comparative Analysis of a Single Global Model Versus a Continent-by-continent Approach

Debris flows, and landslides in general, are worldwide catastrophic phenomena. As world population and urbanization grow in magnitude and geographic coverage, the need exists to extend focus, research, and modeling to a continental and global scale.Although debris flow behavior and parameters are local phenomena, sound generalizations can be applied to debris flow susceptibility analyses at larger geographic extents based on these criteria. The focus of this research is to develop a global debris flow susceptibility map by modeling at both a continental scale for all continents and by a single global model and determine whether a global model adequately represents each continent. Probability Density, Conditional Probability, Certainty Factor, Frequency Ratio, and Maximum Entropy statistical models were developed and evaluated for best model performance using fourteen environmental factors generally accepted as the most appropriate debris flow predisposing factors. Global models and models for each continent were then developed and evaluated against verification data. The comparative analysis demonstrates that a single global model performs comparably or better than individual continental models for a majority of the continents, resulting in a debris flow susceptibility map of the world useful in international planning, and future debris flow susceptibility modeling for determining societal impacts.

Impact study of integrated precipitable water estimated from Indian GPS measurements

The Global Positioning System – Integrated Precipitable Water (IPW) data from Indian stations namely Chennai, Guwahati, Kolkata, Mumbai and New Delhi have been assimilated in the National Centre for Medium Range Weather Forecasting’s (NCMRWF) Global Data Assimilation System (GDAS). Gridpoint Statistical Interpolation (GSI) Scheme of GDAS analysis is experimented with the global model T254L64. The analyses and forecasts are carried out at triangular truncation of wave number 254 and with 64 levels in vertical. Global analyses are carried four times (0000 UTC, 0600 UTC, 1200 UTC and 1800 UTC) daily with intermittent time scheme. Model integrations are carried up to 168 hours. The present study examines the impact that integrated precipitable water has over various meteorological parameters. The study reveals that the assimilation of IPW data influences the analyses and corresponding forecasts of the weather model T254L64. This is an attempt of assimilation of IPW data of the aforesaid five Indian stations in the global model and examination of corresponding impact on various meteorological parameters over Indian region. It is seen that for the layers above 750 hPa the zonal and meridional wind components for IPW analyses have less biases. Forecasts from IPW simulations are found to have consistently by lower 850 hPa wind vector root mean square error (RMSE) where as at 250 hPa, improvement in IPW runs are seen only for day-1 and day-4 forecasts. For temperature at 850 hPa, IPW forecasts valid for day-4 & day-5 are better. At 250 hPa, temperature RMSE for IPW runs is lower for day-1 forecasts. Mean error of IPW forecasts at 250 hPa is lower for all the days of forecasts. Also, geo-potential RMSE for the IPW runs at 250 hPa is lower for all the days of forecasts. Forecasts vs analyses study shows positive impact of IPW assimilation on the anomaly and pattern correlations.

Kinetic study of key species and reactions of atmospheric pressure pulsed corona discharge in humid air

The traditional corona discharge fluid model considers only electrons, positive and negative ions, and the discharge parameters are determined using the simplified weighting method involving the partial pressure ratio. Atmospheric pressure discharge plasma in humid air involves three main neutral gas molecule types: N2, O2, and H2O(g). However, in these conditions, the discharge process involves many types of particles and chemical reactions, and the charge and substance transfer processes are complex. At present, the databases of plasma chemical reaction equations are still expanding based on scholarly research. In this study, we examined the key particles and chemical reactions that substantially influence plasma characteristics. In summarizing the chemical reaction model for the discharge process of N2–O2–H2O(g) mixed gases, 65 particle types and 673 chemical reactions were investigated. On this basis, a global model of atmospheric pressure humid air discharge plasma was developed, with a focus on the variation of charged particles densities and chemical reaction rates with time under the excitation of a 0–200 Td pulsed electric field. Particles with a density greater than 1% of the electron density were classified as key particles. For such particles, the top ranking generation or consumption reactions (i.e., where the sum of their rates was greater than 95% of the total rate of the generation or consumption reactions) were classified as key chemical reactions On the basis of the key particles and reactions identified, a simplified global model was derived. A comparison of the global model with the simplified global model in terms of the model parameters, particle densities, reaction rates (with time), and calculation efficiencies demonstrated that both models can adequately identify the key particles and chemical reactions reflecting the chemical process of atmospheric pressure discharge plasma in humid air. Thus, by analyzing the key particles and chemical reaction pathways, the charge and substance transfer mechanism of atmospheric pressure pulse discharge plasma in humid air was revealed, and the mechanism underlying water vapor molecules’ influence on atmospheric pressure air discharge was elucidated.

An Intelligent Diagnosis Method for Machine Fault Based on Federated Learning

In engineering, the fault data unevenly distribute and difficultly share, which causes that the existing fault diagnosis methods cannot recognize the newly added fault types. An intelligent diagnosis method for machine fault is proposed based on federated learning. Firstly, the local fault diagnosis models diagnosing the existing fault data and the newly added fault data are established by deep convolutional neural network. Then, the weight parameters of local models are fused into global model parameters by federated learning. Finally, the global model parameters are transmitted to each local model. Therefore, each local model update into a global shared model which can recognize the newly added fault types. The proposed method is verified by bearing data. Compared with the traditional model, which can only diagnose the existing fault data but cannot recognize newly added fault types, the federated fault diagnosis model fusing weight parameters can diagnose newly added faults without exchanging the data, and the accuracy is 100%. The proposed method provides an effective method to solve the poor sharing of fault data and poor generalization of fault diagnosis model for mechanical equipment.

Personalized Federated Learning with Semisupervised Distillation

Heterogeneous data and models pose critical challenges for federated learning. However, the traditional federated learning framework, which trains the global model by transferring model parameters, has major limitations; it requires that all participants have the same training model architectures, and the trained global model does not guarantee accurate projections for participants’ personal data. To solve this problem, we propose a new federal framework named personalized federated learning with semisupervised distillation (pFedSD), which ensures the privacy of the participants’ model architectures and improves the communication efficiency by transmitting the model’s predicted class distribution rather than model parameters. First, the server adopts the adaptive aggregation method to reduce the weight of low-quality model predictions for the model’s predicted class distributions uploaded by all clients, which helps to improve the quality of the aggregation of the prediction class distribution. Then, the server sends it back to the clients for local training to obtain the personalized model. We finally conducted experiments on different datasets (MNIST, FMNIST, and CIFAR10), and the results show that the model performance of pFedSD exceeds the latest federated distillation algorithms.