In recent years, subdomain adaptation (SDA) has been widely used to predict the remaining useful life (RUL) of bearings under different working conditions. Although SDA-based methods have achieved accurate predictions, most are still facing challenges, such as fixed subdomain boundaries, the assumption of symmetric domain structures, and ignoring a time-weighted mechanism. These limitations hinder further improvements in prediction accuracy. To address these issues, a subdomain adaptation and contextual information compensation (SDA-CIC) network was proposed that dynamically and adaptively divided subdomains in each training epoch to achieve precise SDA. To validate its effectiveness, hyperparameter analysis, ablation experiments, comparison experiments, and generalization experiments were performed. Experimental results show that the proposed SDA-CIC outperforms the state-of-the-art models.

An issue of great concern in the maintenance of critical infrastructures is the timely and accurate detection of mechanical faults, before these faults deteriorate into failures, causing major service disruptions. The ubiquity of rolling-element bearings (REBs) provides much incentives for online diagnosis of REB faults, spurring a crescendo of research activities that started decades ago. Near-perfect accuracies of state-of-the-art (SOTA) classifiers leveraging deep learning (DL) on the public-domain Case Western Reserve University (CWRU) dataset have been reported. However, the faults covered by this dataset were artificially induced in a controlled environment to bypass the class imbalance problem, and do not include cage faults. Among the open questions to be addressed are: How reproducible are SOTA performances? What are the associated resource requirements? How transferable is SOTA classification performance from CWRU data to actual field data covering naturally occurring faults? How well do SOTA classifiers perform when trained on this real-world dataset, without and with mitigation of the class imbalance problem? How generalisable is classifier performance in the real-world context? Set out to answer the preceding questions, the experimental study presented here provides (i) benchmark results on the CWRU dataset and a real-world dataset from South Australian Water Corporation (SA Water), and (ii) derived from the benchmark results, new insights regarding the state of the research field as well as an answer to the question in the title of this paper: is the problem of REB fault diagnosis a solved problem?

Deep neural networks (DNNs) have become essential for processing the vast amounts of aerial imagery collected using earth-observing satellite platforms. However, DNNs are vulnerable towards adversarial examples, and it is expected that this weakness also plagues DNNs for aerial imagery. In this work, we demonstrate one of the first efforts on physical adversarial attacks on aerial imagery, whereby adversarial patches were optimised, fabricated and installed on or near target objects (cars) to significantly reduce the efficacy of an object detector applied on overhead images. Physical adversarial attacks on aerial images, particularly those captured from satellite platforms, are challenged by atmospheric factors (lighting, weather, seasons) and the distance between the observer and target. To investigate the effects of these challenges, we devised novel experiments and metrics to evaluate the efficacy of physical adversarial attacks against object detectors in aerial scenes. Our results indicate the palpable threat posed by physical adversarial attacks towards DNNs for processing satellite imagery1.

Differential received signal strength (DRSS) provides a practical means of localisation for wireless sensor networks. Closed-form location estimators based on a linearised propagation path loss model are computationally efficient and hence suitable for wireless sensor nodes. However, most existing solutions suffer from a significant bias arising from the injection of noise into the measurement data matrix caused by the linearisation process. The instrumental variable (IV) method can reduce the bias by replacing the data matrix with a matrix that is weakly correlated with the noise and strongly correlated with the data matrix - the latter correlation however can be weakened by large noise. This paper addresses the bias problem when the noise is large by using the method of selective power measurement to maintain correlation between the instrumental variable and data matrix. Simulation results show the proposed solution out-performs other existing solutions in terms of the root mean square error (RMSE) and bias over a wide range of noise levels.

In privacy-preserving machine learning, individual parties are reluctant to share their sensitive training data due to privacy concerns. Even the trained model parameters or prediction can pose serious privacy leakage. To address these problems, we demonstrate a generally applicable Distributed Privacy-Preserving Prediction (DPPP) framework, in which instead of sharing more sensitive data or model parameters, an untrusted aggregator combines only multiple models' predictions under provable privacy guarantee. Our framework integrates two main techniques to guarantee individual privacy. First, we introduce the improved Binomial Mechanism and Discrete Gaussian Mechanism to achieve distributed differential privacy. Second, we utilize homomorphic encryption to ensure that the aggregator learns nothing but the noisy aggregated prediction. Experimental results demonstrate that our framework has comparable performance to the non-private frameworks and delivers better results than the local differentially private framework and standalone framework.

Advancements in deep neural networks have contributed to near-perfect results for many computer vision problems, such as object recognition, face recognition, and pose estimation. However, human action recognition is still far from human-level performance. Owing to the articulated nature of the human body, it is challenging to detect an action from multiple viewpoints, particularly from an aerial viewpoint. This is further compounded by a scarcity of datasets that cover multiple viewpoints of actions. To fill this gap and enable research in wider application areas, in this article we present a multiviewpoint outdoor action recognition dataset collected from YouTube and our own drone. The dataset consists of 20 dynamic human action classes, 2324 video clips, and 503 086 frames. All videos are cropped and resized to 720 × 720 without distorting the original aspect ratio of the human subjects in videos. This dataset should be useful to many research areas, including action recognition, surveillance, and situational awareness. We evaluate the dataset with a two-stream convolutional neural network architecture coupled with a recently proposed temporal pooling scheme called kernelized rank pooling that produces nonlinear feature subspace representations. The overall baseline action recognition accuracy is 74.0%.

The proliferation of wearable devices has contributed to the emergence of mobile crowdsensing, which leverages the power of the crowd to collect and report data to a third party for large-scale sensing and collaborative learning. However, since the third party may not be honest, privacy poses a major concern. In this paper, we address this concern with a two-stage privacy-preserving scheme called RG-RP: the first stage is designed to mitigate maximum a posteriori (MAP) estimation attacks by perturbing each participant's data through a nonlinear function called repeated Gompertz (RG); while the second stage aims to maintain accuracy and reduce transmission energy by projecting high-dimensional data to a lower dimension, using a row-orthogonal random projection (RP) matrix. The proposed RG-RP scheme delivers better recovery resistance to MAP estimation attacks than most state-of-the-art techniques on both synthetic and real-world datasets. For collaborative learning, we proposed a novel LSTM-CNN model combining the merits of Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN). Our experiments on two representative movement datasets captured by wearable sensors demonstrate that the proposed LSTM-CNN model outperforms standalone LSTM, CNN and Deep Belief Network. Together, RG+RP and LSTM-CNN provide a privacy-preserving collaborative learning framework that is both accurate and privacy-preserving.

Multi-hop over-the-air reprogramming is essential for remote installation of software patches and upgrades in wireless sensor networks (WSNs). Several recent coding-based reprogramming protocols have been proposed to enable efficient code dissemination in high packet loss environments. An accurate and formal analysis of the performance of these protocols, however, has not been studied sufficiently in the literature. In this paper, we present a novel high-fidelity analytical model based on the shortest path algorithm to measure the completion time by incorporating overhearing and packet coding. This model can be applied to any coding-based reprogramming protocol by substituting the coding part with protocol specific operations. We conduct extensive testbed experiments to evaluate the performance of our proposed model. Based on the analytical and numerical experiments, we find that 1) overhearing causes significant reduction of the completion time in dense wireless sensor networks, particularly, it reduces 50-70 percent of the total completion time when the packet reception rate is 0.896; 2) coding delay plays a key role in the total completion time compared to the communication delay when the packet coding parameters are selected appropriately, for example, the communication delay is about 65 percent of the coding delay when the number of packets per page is 16 for the finite field size 28; 3) the total completion time can be minimized when the number of packets per page is close to 24 and the finite field size is close to 24.

Multicasting refers to the transmission of a message to multiple receivers at the same time. To enable authentication of sporadic multicast messages, a conventional digital signature scheme is appropriate. To enable authentication of a multicast data stream, however, an authenticated multicast or multicast authentication (MA) scheme is necessary. An MA scheme can be constructed from a conventional digital signature scheme or a multiple-time signature (MTS) scheme. A number of MTS-based MA schemes have been proposed over the years. Here, we formally analyze four MA schemes, namely BiBa, TV-HORS, SCU+ and TSV+. Among these MA schemes, SCU+ is an MA scheme we constructed from an MTS scheme designed for secure code update, and TSV+ is our patched version of TSV, an MA scheme which we show to be vulnerable. Based on our simulation-validated analysis, which complements and at places rectifies or improves existing analyses, we compare the schemes' computational and communication efficiencies relative to their security levels. For numerical comparison of the schemes, we use parameters relevant for a smart (power) grid component called wide-area measurement system. Our comparison shows that TV-HORS, while algorithmically unsophisticated and not the best performer in all categories, is the most balanced performer. SCU+, TSV+ and by implication the schemes from which they are extended do not offer clear advantages over BiBa, the oldest among the schemes.

Multi-hop over-the-air reprogramming is essential for the remote installation of software patches and upgrades in wireless sensor networks (WSNs). Recently, coding-based reprogramming protocols are proposed to address efficient code dissemination in environments with high packet loss rate. The problem of analyzing the performance of these protocols, however, has not been explored in the literature. In this paper, we present a high-fidelity analytical model based on Dijkstra's shortest path algorithm to measure the completion time of coding-based reprogramming protocols. Our model takes into account not only page pipelining and negotiation, but also coding computation. Results from extensive simulations of a representative coding-based reprogramming protocol called Rateless Deluge are in good agreement with the performance predicted by our model, thus validating our approach. Our analytical results show both the number of packets per page and the finite field size have significant impact on completion time. Most notably, the time overhead of coding computation exceeds that of communication when the number of packets per page is 24 and the finite field size is at least 24.

In a tree-based ZigBee network, ZigBee routers (ZRs) must schedule their beacon transmission time to avoid beacon collisions. The beacon schedule determines packet delivery latency from the end devices to the ZigBee coordinator at the root of the tree. Traditionally, beacon schedules are chosen such that a ZR does not reuse the beacon slots already claimed by its neighbors, or the neighbors of its neighbors. We observe, however, that beacon slots can be reused judiciously, especially when the risk of beacon collision caused by such reuse is low. The advantage of such reuse is that packet delivery latency can be reduced. We formalize our observation by proposing a node-pair classification scheme. Based on this scheme, we can easily assess the risk of slot reuse by a node pair. If the risk is high, slot reuse is disallowed; otherwise, slot reuse is allowed. This forms the essence of our ZigBee-compatible, distributed, risk-aware, probabilistic beacon scheduling algorithm. Simulation results show that on average the proposed algorithm produces a latency only 24 percent of that with conventional method, at the cost of 12 percent reduction in the fraction of associated nodes.

ABSTRACT Key management is the pillar of a security architecture. Body sensor networks (BSNs) pose several challenges–some inherited from wireless sensor networks (WSNs), some unique to themselves–that require a new key management scheme to be tailor‐made. The challenge is taken on, and the result is KALwEN, a new parameterized key management scheme that combines the best‐suited cryptographic techniques in a seamless framework. KALwEN is user‐friendly in the sense that it requires no expert knowledge of a user, and instead only requires a user to follow a simple set of instructions when bootstrapping or extending a network. One of KALwEN's key features is that it allows sensor devices from different manufacturers, which expectedly do not have any pre‐shared secret, to establish secure communications with each other. KALwEN is decentralized, such that it does not rely on the availability of a local processing unit (LPU). KALwEN supports secure global broadcast, local broadcast, and local (neighbor‐to‐neighbor) unicast, while preserving past key secrecy and future key secrecy (FKS). The fact that the cryptographic protocols of KALwEN have been formally verified also makes a convincing case. With both formal verification and experimental evaluation, our results should appeal to theorists and practitioners alike. Copyright © 2010 John Wiley & Sons, Ltd.

Proportional, TCP friendly (minimum potential delay) and max-min fairness are three most commonly used fairness criteria for resource allocation in communication networks. In this paper, we generalize the above fairness criteria in terms of utility and study the resource allocation problem for heterogeneous networks where contending users may have different Quality of Services (QoS) requirements and the utility functions may not necessarily satisfy the strict concavity condition, such as real-time applications. We propose a QoS based flow control algorithm and with different link price feedback mechanisms, utility weighted proportional, TCP friendly and max-min fairness is achieved in this unified approach. In addition, the new algorithm is not only suitable for elastic data traffic, but also capable of handling real-time applications, and therefore it can be treated as an efficient flow control mechanism to provide congestion control and QoS balance for Differentiated Services in the future Internet.

Network reprogramming is a crucial service in wireless sensor networks (WSNs) that relies on epidemic strategy for spreading software updates by just having a local view of the networks. Securing the process of network reprogramming is essential in some certain WSNs applications, state-of-the-art secure network reprogramming protocols for WSNs aim for the efficient source authentication and integrity verification of code image, however, due to the resource constrains of WSNs, existing secure network reprogramming protocols are vulnerable to Denial of Service (DoS) attacks when sensor nodes can be compromised (insider DoS attacks). In this paper, we identify different types of DoS attacks exploiting the epidemic propagation strategies used by Deluge and propose corresponding analysis models to attempt to quantify the cost of these attacks damage. Simulation further shows the impact of insider DoS attacks on network reprogramming in WSNs.

Wireless sensor networks (WSNs) are well suited for environment monitoring. However, some highly specialized sensors (e.g. hydrological sensors) have high power demand, and without due care, they can exhaust the battery supply quickly. Taking measurements with this kind of sensors can also overwhelm the communication resources by far. One way to reduce the power drawn by these high-demand sensors is adaptive sampling, i.e., to skip sampling when data loss is estimated to be low. Here, we present an adaptive sampling algorithm based on the Box-Jenkins approach in time series analysis. To measure the performance of our algorithms, we use the ratio of the reduction factor to root mean square error (RMSE). The rationale of the metric is that the best algorithm is the algorithm that gives the most reduction in the amount of sampling and yet the the smallest RMSE. For the datasets used in our simulations, our algorithm is capable of reducing the amount of sampling by 24% to 49%. For seven out of eight datasets, our algorithm performs better than the best in the literature so far in terms of the reduction/RMSE ratio.

In this paper, we propose a novel dynamic routing strategy called tunnel routing that enables efficient and resilient data forwarding from sensors to a destination in wireless sensor networks. Instead of establishing a single static path, tunnel routing identifies a set of trusted sensor nodes that form a tunnel between the source and the destination. Terminals in the tunnel cooperate with each other to reliably route data and to defend against compromised nodes that may maliciously drop or re-direct packets. To achieve this, not only do sensor terminals forward data, but also they act as support nodes, providing on-demand retransmissions in order to combat channel fading, and initializing multi-path data forwarding towards the destination in case they detect a misbehavior of a forwarding node. Simulation results demonstrate the advantages of the proposed scheme.

A typical wireless sensor node has little protection against radio jamming. The situation becomes worse if energy-efficient jamming can be achieved by exploiting knowledge of the data link layer. Encrypting the packets may help to prevent the jammer from taking actions based on the content of the packets, but the temporal arrangement of the packets induced by the nature of the protocol might unravel patterns that the jammer can take advantage of, even when the packets are encrypted. By looking at the packet interarrival times in three representative MAC protocols, S-MAC, LMAC, and B-MAC, we derive several jamming attacks that allow the jammer to jam S-MAC, LMAC, and B-MAC energy efficiently. The jamming attacks are based on realistic assumptions. The algorithms are described in detail and simulated. The effectiveness and efficiency of the attacks are examined. In addition, we validate our simulation model by comparing its results with measurements obtained from actual implementation on our sensor node prototypes. We show that it takes little effort to implement such effective jammers, making them a realistic threat. Careful analysis of other protocols belonging to the respective categories of S-MAC, LMAC, and B-MAC reveals that those protocols are, to some extent, also susceptible to our attacks. The result of this investigation provides new insights into the security considerations of MAC protocols.

A k-connected wireless sensor network (WSN) allows messages to be routed via one (or more) of at least k node-disjoint paths, so that even if some nodes along one of the paths fail, or are compromised, the other paths can still be used. This is a much desired feature in fault tolerance and security, k-connectivity in this context is largely a well-studied subject. When we apply the random key pre-distribution scheme to secure a WSN however, and only consider the paths consisting entirely of secure (encrypted and/or authenticated) links, we are concerned with the secure k-connectivity of the WSN. This notion of secure k-connectivity is relatively new and no results are yet available. The random key pre-distribution scheme has two important parameters: the key ring size and the key pool size. While it has been determined before the relation between these parameters and 1-connectivity, our work in k-connectivity is new. Using a recently introduced random graph model called kryptograph, we derive mathematical formulae to estimate the asymptotic probability of a WSN being securely k-connected, and the expected secure k-connectivity, as a function of the key ring size and the key pool size. Finally, our theoretical findings are supported by simulation results.

Cryptographic algorithms play an important role in the security architecture of wireless sensor networks (WSNs). Choosing the most storage- and energy-efficient block cipher is essential, due to the facts that these networks are meant to operate without human intervention for a long period of time with little energy supply, and that available storage is scarce on these sensor nodes. However, to our knowledge, no systematic work has been done in this area so far. We construct an evaluation framework in which we first identify the candidates of block ciphers suitable for WSNs, based on existing literature and authoritative recommendations. For evaluating and assessing these candidates, we not only consider the security properties but also the storage- and energy-efficiency of the candidates. Finally, based on the evaluation results, we select the most suitable ciphers for WSNs, namely Skipjack, MISTY1, and Rijndael, depending on the combination of available memory and required security (energy efficiency being implicit). In terms of operation mode, we recommend Output Feedback Mode for pairwise links but Cipher Block Chaining for group communications.

The energy efficiency requirement of wireless sensor networks (WSN) is especially high because the sensor nodes are meant to operate without human intervention for a long period of time with little energy supply. Besides, available storage is scarce due to their small physical size. Therefore choosing the most storage- and energy-efficient block cipher for WSN is important. However to our knowledge so far no systematic work has been conducted in this area. We have identified the candidates of block ciphers suitable for WSN based on existing literature and authoritative recommendations. We have benchmarked these candidates and based on this benchmark, we have selected the suitable ciphers for WSN, namely Rijndael for high security and energy efficiency requirements; but MISTY1 for good storage and energy efficiency. In terms of operation mode, we recommend output feedback mode for static networks, but counter mode for dynamic networks.