Search

In the fields of deep learning and computer vision,the security of object detection models has received extensive attention. Revealing the security vulnerabilities resulting from adversarial attacks has become one of the most important research directions. Existing studies show that object detection models can also be threatened by adversarial examples,just like other deep neural networks based models,e.g.,those for classification. In this paper,we propose a bidirectional adversarial attack method. Firstly,the added perturbation pushes the prediction results given by the object detectors far away from the groundtruth class while getting close to the background class. Secondly,a condence loss function is designed for the region proposal network to reduce the foreground scores. Thirdly,the adversarial examples are generated by a pre-trained autoencoder,and the model is trained using an adversarial approach,which can enhance the similarity between the adversarial examples and the original image and speed up algorithm convergence. The proposed method was verified on the most popular two-stage detection framework (Faster R-CNN),and 55.1% mAP-drop were obtained. In addition,the adversarial examples have superior transferability,migrating which to the common one-stage detection framework (YOLOv3) gets a 39.5% mAP-drop.

Skeletal fluorosis is a form of endemic disease caused by the excessive intake of fluoride. Bone deformation and periosteal calcification are the typical manifestations that can be observed in the images and are usually served as a basis of pathological grading. In the current medical systems, the diagnosis of skeletal fluorosis fully relies on doctors’ knowledge and experience, and no research effort has been made in automatic image information diagnostic systems. According to the image information, the shape of the lesion is irregular, the boundary is unclear and the lesion related pixels only occupy a small part of the image. We take the lead in proposing a two-stage scheme that can achieve automated X-ray image diagnosis and severity grading. In the first stage, a Dense U-Net is proposed for reliable lesion determination, and a multitype feature fusion approach passes effective and comprehensive features to the subsequent stage. In the second stage, a novel classifier is designed with the integration of ensemble learning and multiple instance learning, which can ensure classification accuracy in case that the feature for diagnosis only takes up a small proportion of the whole image. Through plenty of experiments on the actual data collected from the hospitals, it is verified that the proposed strategy can achieve satisfactory results on skeletal fluorosis image diagnosis and severity grading.

In a paced aircraft final assembly line, some disturbances can be collected timely on the basis of the cyber-physical production system. In order to reduce the execution deviation, some workers need to switch among stations after a fixed period. Thus, a worker allocation problem with the multistage workstation is introduced first. Then, an integer programming formulation is presented to formulate the problem with the objective of shortest workstation cycle and the workload balance of both stations and workers. Moreover, a modified nondominated sorting genetic algorithm (NSGA-IV) is proposed to solve it, which tradeoff the convergence and the population diversity in the decision space. Finally, the NSGA-IV algorithm compares with five multiobjective evolutionary algorithms in a real-world case. Compared with manual allocation, the takt time of an aircraft final assembly line is reduced by 20.86% by using the NSGA-IV algorithm.

Over the past twenty years, numerous research outcomes have been published, related to the design and implementation of soft sensors. In modern industrial processes, various types of soft sensors are used, which play essential roles in process monitoring, control and optimization. Emerging new theories, advanced techniques and the information infrastructure have enabled the elevation of the performance of soft sensing. However, novel opportunities are accompanied by novel challenges. This work is motivated by these observations and aims to present a comprehensive review of the developments since the start of the millennium. While a few books and review articles are published on the related topics, more focus on the most up-to-the-date advancement is put in this work, from the perspective of systems and control.

This article proposes a fault-tolerant compensation control approach against nonlinearity, simultaneous additive, and multiplicative actuator faults in Markov jump systems. In this article, we first exploit the fuzzy logic system (FLS) to approximate the nonlinear functions, which have no available knowledge. Then, by utilizing the adaptive backstepping technique, a FLS-based adaptive fault-tolerant compensation controller is proposed, which can completely compensate for the adverse effects, arising from the additive actuator faults, the multiplicative actuator faults, and the mismatched nonlinearity simultaneously. The stability of the closed-loop system can be guaranteed by the proposed FLS-based adaptive controller with the adaptation laws. The novelty of this article lies in the fact that the additive and multiplicative actuator faults, and mismatched nonlinearity are considered simultaneously. Besides, the renown sliding mode control approach has limitations to deal with the FTC problem considered in this article because the considered nonlinearity is a mismatched one. The proposed control approach can cope with the challenging case. Finally, a practical wheeled mobile manipulator system is used to demonstrate the effectiveness and validity of the proposed approach.

Retinal vessel image is an important biological information that can be used for personal identification in the social security domain, and for disease diagnosis in the medical domain. While automatic vessel image segmentation is essential, it is also a challenging task because the retinal vessels have complex topological structures, and the retinal vessels vary in size and shape. In recent years, image segmentation based on the deep learning technique has become a mainstream method. Unfortunately, the existing methods cannot make the best use of the global information, and the model complexity is high. In this article, a convolutional neural network integrated with the attention mechanism is proposed. The overall network structure consists of a basic U-Net and an attention module, and the latter is used to capture global information and to enhance features by placing it in the process of feature fusion. Experiment results on five public datasets show that the proposed scheme outperforms other existing mainstream approaches, and most of the performance indicators are in the leading positions. More importantly, the proposed method has a significant reduction in the number of parameters.

In the conventional approaches to the design of fault diagnosis systems, little effort is usually paid to the selection of the parity vectors. As a result, the systems’ performance can be significantly affected. In this article, novel approaches are proposed to derive the parity vectors that construct optimized residual generators for linear and nonlinear systems. Based on the analysis on the parity space dimension, a novel parameterization of all parity relation-based residual generators is proposed. An iterative procedure that guarantees minimal regression error is then employed in the search for the optimal parameters. Considering that the traditional parity relation-based approaches are only suitable for linear systems, in this work, the proposed approach is also generalized to deal with strong nonlinearities, with the aid of data-driven Hammerstein function estimation. Furthermore, optimized residual generation algorithms are summarized for offline design and online implementation, the performance of which is evaluated thoroughly with a three-tank system, a numerical nonlinear example, as well as a case study on an industrial hot rolling mill process. Results show that residuals generated by the proposed approaches can significantly improve the sensitivity to small faults, and thus, the fault detection rate is improved compared with the traditional nonoptimized approach.

Data centers are fundamental facilities that support high-performance computing and large-scale data processing. To guarantee that a data center can provide excellent properties of expanding and routing, the interconnection network of a data center should be designed elaborately. Herein, we propose a novel structure for the interconnection network of data centers that can be expanded with a variable coefficient, also known as a variable expanding structure (VES). A VES is designed in a hierarchical manner and built iteratively. A VES can include hundreds of thousands and millions of servers with only a few layers. Meanwhile, a VES has an extremely short diameter, which implies better performance on routing between every pair of servers. Furthermore, we design an address space for the servers and switches in a VES. In addition, we propose a construction algorithm and routing algorithm associated with the address space. The results and analysis of simulations verify that the expanding rate of a VES depends on three factors:n,m, andkwhere thenis the number of ports on a switch, themis the expanding speed and thekis the number of layers. However, the factormyields the optimal effect. Hence, a VES can be designed with factormto achieve the expected expanding rate and server scale based on the initial planning objectives.

The intensive research and development efforts directed towards large-scale complex industrial systems in the context of Industry 4.0 indicate that safety and reliability issues pose significant challenges. During online operation, system performance degradation will lead, not only to economic losses, but also potential safety hazards. In the existing research and technical routes, the target of the fault diagnosis systems is to trigger alarms to report the fact of the existence of malfunctions as well as the underlying reasons accurately. However, it remains unanswered how urgent it is to fix it, and what degrees of fault-tolerance, maintenance, and fault recovery are needed. Further analyses are necessary to evaluate the impact of the detected fault on the plant-wide performance. In this article, to enable a more comprehensive and precise description of the plant-wide operational status, the roles of the commonly used performance metrics, the state-of-the-art performance evaluation approaches, as well as the performance-oriented and plant-wide process monitoring techniques are investigated. On this basis, an alternative straightforward technical route, embedded in the cyber-physical-social system framework is proposed. A roadmap including the key research questions, the future research directions, and an outlook about the future vision is presented.

With the wide integration of the Cyber-Physical System (CPS) and Internet of things (IoT), the manufacturing industry has entered into an era of big data. Thus, manufacturing companies are facing challenges when conducting Big Data Analytics, including the high velocity of data generation, the enormous volume, the multifarious formats and types as well as the quality or fidelity. In this paper, a Cyber-Physical Production System (CPPS) using data analytics is proposed to enable production visibility. Firstly, this study uses data stream processing approaches to clean redundant data efficiently. Secondly, a Bayesian inference engine, which is trained by ming the historical data offline, is employed to identify the accuracy of an RFID-captured event online. Then, complex event processing is applied to fuse multi-source heterogeneous data. Finally, production progress visibility is achieved by the Business Process Management. The proposed system demonstrates that it is significant to implement real-time data collection, processing and visibility, as well as to improve production efficiency. A demonstrative case from the machinery industry is presented to validate the CPPS.

Recommendation systems have been widely used for user data mining. However, considering the security of user privacy and legal regulations, various institutions have begun to strengthen data privacy protection gradually. Federated learning is a novel joint training privacy protection framework, which can be executed without knowing the original data of each participant. In this paper, we propose a method based on clustering and latent factor model under the vertical federated recommendation system. Taking into account the diversity of a large number of different users in each participant and the complexity of the matrix factorization of the user-item matrix, we cluster the users to reduce the dimension of the matrix and improve the accuracy of user recommendations. In addition, in order to ensure the security of user data privacy, we use homomorphic encryption to protect data. We conducted experiments on the MovieLens movie dataset. Experiments show that effective clustering can significantly improve the accuracy of the recommendation system of each participant.

Deep learning techniques have promoted the rise of artificial intelligence (AI) and performed well in computer vision. Medical image analysis is an important application of deep learning, which is expected to greatly reduce the workload of doctors, contributing to more sustainable health systems. However, most current AI methods for medical image analysis are based on supervised learning, which requires a lot of annotated data. The number of medical images available is usually small and the acquisition of medical image annotations is an expensive process. Generative adversarial network (GAN), an unsupervised method that has become very popular in recent years, can simulate the distribution of real data and reconstruct approximate real data. GAN opens some exciting new ways for medical image generation, expanding the number of medical images available for deep learning methods. Generated data can solve the problem of insufficient data or imbalanced data categories. Adversarial training is another contribution of GAN to medical imaging that has been applied to many tasks, such as classification, segmentation, or detection. This paper investigates the research status of GAN in medical images and analyzes several GAN methods commonly applied in this area. The study addresses GAN application for both medical image synthesis and adversarial learning for other medical image tasks. The open challenges and future research directions are also discussed.

We present a novel, easy-to-use method based on the photon-mapping technique to simulate photometric images of moving targets. Realistic images can be created in two passes: photon tracing and image rendering. The nature of light sources, tracking mode of the telescope, point spread function (PSF), and specifications of the CCD are taken into account in the imaging process. Photometric images in a variety of observation scenarios can be generated flexibly. We compared the simulated images with the observed ones. The residuals between them are negligible, and the correlation coefficients between them are high, with a median of $0.9379_{-0.0201}^{+0.0125}$ for 1020 pairs of images, which means a high fidelity and similarity. The method is versatile and can be used to plan future photometry of moving targets, interpret existing observations, and provide test images for image processing algorithms., 17 pages, 7 figures

Motivated by the importance of service quality in today's customer business environment, we consider two periodic-review stochastic inventory models with probabilistic service-level guarantees for restricting stockout probabilities: (i) the classical inventory control model with backlogging and (ii) the remanufacturing inventory control model with random product returns. We assume that demands are stochastic nonstationary, evolving and correlated over time. We establish the optimality of generalized base-stock policies, and propose new 2-approximation algorithms for both models. The core concept developed in this paper is called the delayed forced holding and production cost, which is proven effective in dealing with service-level constrained inventory systems. Our extensive computational experiments show that the proposed algorithms on average perform within 2% error of optimality.

Process safety, system reliability, and product quality are becoming increasingly essential in the modern industry. As a result, prognosis and fault diagnosis of the complex systems have gained a substantial amount of research attention. In order to evaluate the influence of the detected faults to systems’ behavior, there is a pressing need to design prognosis and diagnosis systems oriented to the key-performance-indicators (KPIs). Dedicated to this requirement, we have recently developed a MATLAB toolbox data based key-performance-indicator oriented fault detection toolbox (DB-KIT), which realizes a series of effective algorithms, to provide a systematic and illustrative material to the peer researchers. This paper investigates the recent advances in the multivariate statistical analysis based approaches. Formulations based on the optimization problems are proposed to better clarify the ideas behind different solutions and to study them in a unified data-driven framework. Theoretical fundamentals of some selected algorithms in the DB-KIT are elaborated. Moreover, new evaluation results on dataset defects are presented, which compare the algorithms’ robustness and demonstrate the power of DB-KIT. The open-source code and the demonstrative simulations can be regarded as baseline and resources for innovation research, comparative studies, and educational purposes.

In order to support China’s major demand of constructing ‘Manufacturing Power’ and creating ‘Advanced Manufacturing Research and Development Base’, Tianhe Industrial Cloud Platform (TICP) has comprehensively broken through the technical challenges such as high performance, high integration and high availability of information platform. TICP is a powerful genetic cloud-based platform as it is developed via the integration of cloud platform and high-performance computing (HPC), which means that TICP have the advantages of cloud computing and supercomputing. TICP breaks through the difficulties of supercomputing applied in advanced manufacturing based on Tianhe-1A. Tianhe-1A, the fastest computer in the world from October 2010 to June 2011, is one of the few Peta scale supercomputers in the world. By utilizing Tianhe-1A, TICP is generic as it supports a wide range of applications while keeping high-level efficiency, stability, security and good user experience. TICP, a collaborative cloud platform driven by supercomputing, has been approved as one of the first batch of ‘Industrial Cloud’ in China and became a national ‘Service-Oriented Manufacturing Demonstration Platform’. The major objectives of the TICP are (1) improving the research and development efficiency of small and medium-sized enterprises, (2) cultivating new economic growth points and (3) gathering international technical teams. TICP is one of the most heat focused industrial application cases that cultivated new forms of business and laid a solid foundation for the construction of Industrial Internet Platform.

This paper focuses on the fault-tolerant control problem of Markov jump systems (MJS) with $It\hat{o}$ stochastic process and output disturbances. Such a problem widely exists in practical systems such as mobile manipulator systems. Since MJS can suitably describe mobile manipulator systems, in this paper, a new approach based on the MJS model is proposed. First, a proportional-derivative sliding mode observer (SMO) and an observer-based controller are designed and synthesized. Two new theorems are derived to ensure the close-loop stochastic stability and the reachability of the sliding mode surface. Compared with the existing works, the system model is more general, which could describe a larger variety of plants or processes. The controller design procedure is simplified by solving the sliding mode parameters and the controller gain simultaneously with only one linear matrix inequality problem. In addition, the augmented fault vector can be reconstructed by employing a descriptor SMO. Simulations are provided to demonstrate the validity of the derived theorems and the effectiveness of the proposed algorithm.

Bearings play an irreplaceable role in various categories of high-precision mechanical equipment. To reduce the adverse impact caused by sudden bearing failures, this paper proposes a method based on improved particle swarm optimization (IPSO), empirical mode decomposition (EMD), and support vector regression (SVR) to predict the remaining useful life (RUL) of bearing. The EMD method decomposes the raw vibration signal, and the components are chosen out according to the kurtosis criterion to reconstruct the vibration signal. After filtering out the noise, the features in time-domain and frequency-domain are extracted from the reconstructed vibration signal. The IPSO method is used to optimize three important parameters of the SVR model. Then the SVR model is trained with the extracted features to predict the bearing RUL. Actual bearing failure dataset is used to verify the validity of this method. Three different methods are used to compare with this proposed method. The results indicate that this proposed method performs best among the four methods in different evaluation indexes. The proposed method makes sense for bearing failure prediction.

The normal operation of rotating machineries depends on the health conditions of rolling element bearings. Once bearings fail, it will cause economic loss and even threaten operational safety. Therefore, it is essential to evaluate the health status of the bearings, where predicting the remaining useful life (RUL) is a quantitative evaluation method. To enable to learn from small-scale datasets of degraded bearings for RUL prediction, this work proposes to construct a support vector regression (SVR) prediction model based on Bayesian optimization (BO). An improved approach based on the $3\sigma$ interval was put forward to determine an optimal time to start prediction. The proposed method is verified on the bearing datasets in the IEEE PHM 2012 challenge. Experiment results verified that the time to start prediction is essential to building an accurate degradation model. Moreover, the BO algorithm is demonstrated to be superior in the optimization of SVR hyper-parameters, especially for low-dimensional data.

Malicious external attackers have been targeting the vulnerabilities in the network-involved industrial control systems such as the smart grids. The happening of several real-world attack events and the possibility of concealed integrity attacks revealed by recent literature are fostering the safety and security experts to pay attention to the dependability of the measurement data after network transmission. Aiming at this, this paper proposes a lightweight data masking/encryption scheme to protect the industrial measurements from being eaves-dropped during network transmissions. At the receiver's end, the measurements are reconstructed using an offline designed decryptor. These data are dependable for subsequent uses in the condition that the attack detection indicators are below the threshold. This proposed scheme is theoretically supported by the correlation relationship between the variables. The confidentiality can be guaranteed because an essential part of the information for solving the reconstruction matrix is never accessible to the attackers. Simulation results on a two-area load frequency control power grid system demonstrate the correctness of the theories and show the applicability of the proposed scheme.

Amidst rising rates of adolescent obesity and overuse of digital technologies amongst secondary school students, educators and policymakers are especially keen on boosting sports participation amongst Chinese youth. Furthermore, the existence of a robust relationship between sports participation and academic performance would be an additional reason to expand sports-related activities in school. The Ministry of Education of China is also calling for attention for Physical Education in recent years. In statistics, ordinary least squares (OLS) is a type of linear least squares method for estimating the unknown parameters in a linear regression model. It is widely used in social sciences as it is easy to understand and calculate. In fact, numerous studies have found that sports participation does have a positive effect on students’ academic performance based on regression model. By using the OLS and interaction effect on the data of China Education Panel Survey (CEPS) between the 2013 - 2014 academic years, this paper use data study correlations between sports participation and academic performance. Based on a sample size of 4033 secondary school students from China, this paper shows that there is a negative correlation between students’ sports intensity and exam scores. Also, this study found that sex plays an interaction effect on this negative correlation. Findings of this paper indicate that the lack of a comprehensive and robust Physical Education program is a weakness of the current Chinese education.

Insulated gate bipolar transistor (IGBT) has important applications in industrial development. However, the IGBT has a complex integrated structure and works in a harsh environment, so it is prone to fault and therefore causes economic losses. Considering this, it is of great significance to use data-driven fault prediction. The fault prediction of IGBT aging parameters based on device data can avoid complex modeling procedure and the workload, so it has wider applicability. In this paper, an adaptive boosting approach based on a multi-class neural network is proposed to realize the analysis of the fault prediction of the aging parameters of IGBTs. This paper uses the IGBT accelerated aging test data released by NASA for processing. In order to improve the quality of data processing, exponentially weighted moving average (EWMA) and outlier processing are used to preprocess. Various neural network approaches are used for time series prediction. Finally, the adaptive boosting algorithm based on single-class and multi-class neural networks is used to achieve better prediction performance compared to the neural network algorithms. The results show that the adaptive boosting approach to integrating multi-class neural networks has a good prediction performance.

This brief proposes a subspace-aided fault detection approach for the drive systems of strip rolling mills. Considering the impact of the unknown periodic load generated by the strip rolling process, the primary contributions are concluded as follows. First, this brief presents an approach to describe the subspace of the unknown/unmeasurable periodic load. Second, a fundamental frequency identification approach for the drive systems is proposed and then the subspace of the unknown periodic load can be constructed by the fundamental frequency. Third, this brief presents a subspace-aided fault detection approach to identify the data-driven stable kernel representation (SKR) of the closed-loop system by projecting the input-output (I/O) process data, so as to obtain a robust residual against the unknown periodic load. In addition, the effectiveness and performance of the approaches are verified by numerical examples and experimental data of the test rig for the drive systems of strip rolling mills. The results show that a robust residual generation against the unknown periodic load in the drive systems can be obtained and the robust subspace-aided fault detection can be achieved. Compared with the traditional method, the proposed approaches can improve the fault detection rate more effectively and be applied to the drive systems of strip rolling mills reliably. © 2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

The fourth industrial revolution is elevating the overall flexibility and controllability of the production processes, leading to better consistency in quality and improved feasibility of personalization. However, an evident downside is observed from the boosting system openness because the security and safety protocols are not ready for the potential impacts and threats. In this work, an in-depth analysis about the novel challenges is firstly presented. Then, the mechanism of false data injection attack on the industrial cyber-physical systems is studied. The target is to reveal the condition when the existing fault diagnosis systems fail to detect the false data injection attack on the distributed and networked systems. A theorem is proposed to support the discussions. It is elaborated that establishing interconnections between the subsystems will help prevent undetectable attacks, and thus improving the systems’ safety.

The safety, stability and reliability of the modern complex processes have always been the focus of the industry. An abnormity can lead to failures in the production and manufacturing processes or even dramatic accidents. The fault diagnosis techniques aim to enhance the aforementioned aspects by detecting the system’s deviations from the normal operating conditions and providing early warnings. By mining the hidden system features in the historical data, complex physical modeling procedures and the dependence on large amounts of prior knowledge can be avoided. In many practical scenarios, data-driven fault diagnosis algorithms are more suitable for modern industrial diagnosis. In this paper, a novel approach is proposed which integrates both multivariate statistical analysis and deep neural network to deal with the nonlinearities in the complex systems. Based on the theory of traditional data-driven methods, deep learning methods and the newly proposed method, a MATLAB-based fault diagnosis toolbox is developed and published online. Plentiful function libraries are provided to the researchers to analyze those algorithms and satisfy the need of practical industrial applications. By applying the developed toolbox, the characteristics of those algorithms are also compared, especially on the time-consumption feature and the fault discrimination feature.

Cyber-Physical Systems (CPSs), belonging to the interdisciplinary field of research, which has spurred broad interests, are always suffering from malicious actuator attacks. In this paper, we propose a fuzzy logic system-based adaptive controller for CPSs under state-dependent actuator attacks. The FLS is utilized to approximate the nonlinear functions related to the state-dependent actuator attacks. By employing adaptive backstepping method, an adaptive controller is developed to compensate the actuator attacks. With the proposed adaptive controller, the uniform ultimate boundedness of the closed-loop plant can be guaranteed. Finally, a numerical example in the face of actuator attacks is employed to illustrate the effectiveness of the proposed method.

To analyze the biomechanical effect of syndesmotic screw through three and four cortices, a total of 12 finite element models simulating healthy ankles, tibiofibular syndesmosis injured ankles, and post-operative ankles by screw fixations through three or four cortices were built. A set of biomechanical data were obtained to find that screw fixation methods for inferior tibiofibular syndesmosis can help recover most of the biomechanical relations of the ankle especially the tricortical fixation, while the screw of quadricortical fixation bear more stress than the tricortical fixation. The modeling method for finite element models was also modified for saving more time and realizing personalized modeling for clinical application.

Industrial cyber-physical systems (ICPSs) are the backbones of Industry 4.0 and as such, have become a core transdisciplinary area of research, both in industry and academia. New challenges brought about by the growing scale and complexity of systems, insufficient information exchange, and the exploitation of knowledge available have started threatening the overall system safety and stability. This work is motivated by these challenges and the strategic and practical demands of developing ICPSs for safety-critical systems such as the intelligent factory and the smart grid. It investigates the current status of research in ICPS monitoring and control, and reviews the recent advances in monitoring, fault diagnosis, and control approaches based on data-driven realization, which can take full advantage of the abundant data available from past observations and those collected online in real time. The practical requirements in the typical ICPS applications are summarized as the major issues to be addressed for the monitoring and the safety control tasks. The key challenges and the research directions are proposed as references to the future work.

A digital twin (DT) is classically defined as the virtual replica of a real-world product, system, being, communities, even cities that are continuously updated with data from its physical counterpart, as well as its environment. It bridges the virtual cyberspace with the physical entities and, as such, is considered to be the pillar of Industry 4.0 and the innovation backbone of the future. A DT is created and used throughout the whole life cycle of the entity it replicates, from cradle to grave, so to speak. This article focuses on the present state of the art of DTs, concentrating on the use of DTs in industry in the context of smart manufacturing, especially from the point of view of plantwide optimization. The main capabilities of DTs (mirroring, shadowing and threading) are discussed in this context. The article concludes with a perspective on the future. This article is part of the theme issue ‘Towards symbiotic autonomous systems’.

Enabling the models with the ability of reasoning and inference over text is one of the core missions of natural language understanding. Despite deep learning models have shown strong performance on various cross-sentence inference benchmarks, recent work has shown that they are leveraging spurious statistical cues rather than capturing deeper implied relations between pairs of sentences. In this paper, we show that the state-of-the-art language encoding models are especially bad at modeling directional relations between sentences by proposing a new evaluation task: Cause-and-Effect relation prediction task. Back by our curated Cause-and-Effect Relation dataset (Cℰℛ), we also demonstrate that a mutual attention mechanism can guide the model to focus on capturing directional relations between sentences when added to existing transformer-based models. Experiment results show that the proposed approach improves the performance on downstream applications, such as the abductive reasoning task.

Cervical cancer is one of the most dangerous diseases among female diseases. Because its clinical manifestation is not obvious, existing diagnostic approaches rely more on regular examination and timely detection of lesions. In this paper, the application of Support Vector Machine (SVM) classifier for classification and diagnosis is firstly introduced, Then a modified approach combining classifier model with partial least squares (PLS) regression is proposed, which can achieve further recognition and diagnosis after eliminating redundant information, so as to obtain a more stable and rapid diagnostic model for screening people at risk of disease. The improved approach and the principal component analysis-support vector machine (PCA-SVM) approach proposed by other researchers are applied to diagnose medical data. In this paper, the effectiveness of the two approaches is verified by simulation tests on cervical cancer data. Compared with existing approaches, results show that the proposed approach has better performance than PCA-SVM in the diagnosis of cervical cancer, and it achieves high classification accuracy by using fewer features.

Aerial images are the images captured from high attitudes above the ground. Processing and analyzing aerial images play central roles in terrain modeling, agricultural monitoring, city planning, environmental surveillance, etc. Aerial images are developing towards high resolution and large size, which poses a major challenge in pixel-level image segmentation. With the rapid development of deep learning technology, the application of deep learning to image semantic segmentation has obtained satisfactory effect. In this paper, we propose a novel aerial image segmentation method based on convolutional neural network (CNN). The main structure of the proposed network adopts U-Net. In order to capture objects of different scales in the deep features, a group of cascaded dilated convolution is inserted at the bottom of U-Net which has different dilation rates. Furthermore, to better optimize the network at different scales, an auxiliary loss function is proposed to be integrated in the cascaded dilated convolution. The effectiveness of the proposed method is evaluated on the Inria Aerial Image Labeling Dataset. Experiment results show that the proposed method has better segmentation performance than existing approaches.

Deep learning technology has greatly improved the performance of target tracking, but most recently developed tracking algorithms are short-term tracking algorithms, which cannot meet the actual engineering needs. Based on the Siamese network structure, this paper proposes a long-term tracking framework with a persistent tracking capability. The global proposal module extends the search area globally through the construction of a feature pyramid. The local regression module is mainly responsible for the confidence evaluation of the candidate regions and for performing more accurate bounding box regression. To improve the discriminative ability of the regression network, the error samples are eliminated by synthesizing the temporal information and are then classified through a verification module in advance. Experiments on the VOT long-term tracking dataset and the UAV20L aerial dataset show that the proposed algorithm achieves state-of-the-art performance.

The cooling system called flowcool is an important part of the ion mill etching (IME) machine. In the case of the cooling system failure during operation, it will lead to significant impacts on the final quality of wafers. To address this problem, effective maintenance plans are made according to the predicted time to failure, i.e., the remaining useful life (RUL). However, the RUL prediction performances of existing prognostic approaches for the flowcool system are not satisfactory. In this work, multiple deep recurrent neural network-based approaches are used for RUL prediction. To improve reliability, the random forest-based early degradation warning approach is employed before the RUL prediction. The long short-term memory (LSTM) neural network, the gated recurrent unit (GRU) neural network, and the additional fully-connected layers (FCs) are used to predict the RUL of flowcool respectively. Results of the comparison study on the dataset of the 2018 PHM Data Challenge competition show that the proposed GRU and GRU-FCs based approaches integrated with sample screening, new feature construction, and degradation warning outperform the other reported approaches. The symmetric mean absolute percentage errors (SMAPE) of the RUL prediction in 3 failure modes achieved 14.13%, 14.44%, and 25.63% respectively.

Chemical mechanical polishing (CMP) is one of the most powerful technologies to achieve global planarization for precision machining of the wafer surface. CMP contributes to intelligent manufacturing in Industry 4.0. Prediction of the material removal rate (MRR) is of vital significance for product quality control during the CMP process. There is no generally accepted theory to expound the principle of material removal in CMP. This paper proposes data-driven approaches to predict MRR based on deep learning methods to pursue better prediction performance. Random forest is employed to obtain the process variables which have a significant influence on MRR prediction and act as the input of the neural network. In addition, it is firstly proposed to predict MRR with the aid of residual convolutional neural network (ResCNN). The dataset provided by the PHM2016 Data Challenge is applied to compare the MRR prediction performance of the ResCNN with the CNN-based approach. Experimental results show that the prediction performance of ResCNN outperforms all the existing approaches reported in the literature.

To solve the problem that the detection accuracy of remote sensing image is affected by convolution neural network overfitting under the condition of small samples, a data augmentation method based on style transfer is proposed, in which new data is generated from texture of external domain to source domain by using cycle-consistent adversarial networks(CycleGAN). The experiment results show that the accuracy of detection and recognition is improved after adding the generated data to the original data.

The imperfect network information security technology restricts the development process of network informatization of power systems. This paper mainly studies the network information security of power systems, analyzes the risks faced by power system network operation, and puts forward some preventive measures for the network information security of power systems.

Aiming at the problem of low accuracy of image classification under the condition of few samples, an improved method based on Wasserstein Generative Adversarial Nets is proposed. The small data sets are augmented by generating target samples through minimax game between the generator and the discriminator. Firstly, label information is introduced into the network to save the cost of manual labeling for the generated images, and then classifier loss is introduced to further improve the quality of generated images, and the classification accuracy is improved after expanding the data set.

This paper focuses on the design of diagnostic observer based residual generator (DORG) for fault detection purposes. The property of the existing Multiple- Input-Single-Output (MISO) DORG is firstly discussed, followed by a parity vector based solution. Then, a novel Multiple- Input- Multiple-Output (MIMO) DORG is proposed through rigorous mathematical derivations. Compared with existing approaches, the proposed approach and algorithms retain the correlation information in the output variables, and reduce the offline design complexity and the online implementation efforts. Simulation studies on a numerical example show that the proposed approach has better fault detection performance than the MISO DORG based approach.

The next generation of industrial revolution aims to achieve flexible and intelligent operation and manufacturing oriented to the customers. In recent years, it has been a core subject to study the large-scale integrated design and implementation of plant equipment based on cyber-physical systems (CPS). However, the expansion of system scales and the exponentially increased complexity also introduce potential safety hazards. Traditional process monitoring and fault diagnosis strategies are no longer suitable for plant-wide management and supervisory tasks. This paper elaborates the role of CPS in the industrial context based on the analysis of the historical development factor and the current condition and trend. A novel framework is proposed for large-scale system online monitoring and diagnosis oriented to plant-wide performance optimization.

This paper presents a dynamic simulation model of tethered-wind power systems, consisting an unmanned aerial vehicle (UAV) and a generator placed on the ground level. The UAV is connected with the generator using a tether, which the tether is oriented upwind and downwind. A simple model was developed in MATLAB using ODE45 solver to capture the dominant dynamics and to verify the feasibility of the system. The simulation shows that the proposed system is viable and capable of periodic motions. The power out of the system was simulated, which predicts the average power out is about 600 watts in a 15m/s wind.

In this paper, a data driven sensor fault tolerant strategy is proposed for nonlinear systems. The core of the proposed strategy is just-in-time learning based soft sensor. When sensor fault occurs, the value of soft sensor is adopted as the redundancy instead of the real faulty sensor value. Meanwhile, the influence of the sensor fault can be tolerated. Due to the complexity of mechanism model for nonlinear system, a kind of just-in-time learning method is employed for soft sensor. The indexes will be predicted by just-in-time learning method online. And only historical sample data will be used for prediction. Instead of considering system global model as normal soft-sensor approaches, just-in-time learning methods only consider the approximate system at current time. Thus JITL owns strong capacity of on-line implementation. Two nonlinear systems, a typical numerical one and a benchmark of wastewater treatment system, are employed for experiments. The experiment results verify the accuracy and implementability of the proposed scheme.

A new MATLAB toolbox DB-KIT was recently developed for the design and implementation of fault diagnosis systems. For the purpose of key performance indicator (KPI) oriented fault detection, over the past few years, a series of test statistics and the corresponding thresholds were derived based on the modified data structures originating from the existing multivariate statistical analysis tools. These data-driven approaches are numerically reliable, efficient and of high fault detection performance. Especially, the false alarm rates (FARs) under KPI-unrelated fault scenarios are suppressed with great efforts, which is the central task of the KPI-oriented fault detection problem. DB-KIT was firstly introduced at the 2016 IEEE Industrial Electronics Conference, and the initial results on algorithm efficiency and fault detection performance were reported in Comparison of KPI related fault detection algorithms using a newly developed MATLAB toolbox: DB-KIT with simulation tests on the Tennessee Eastman Process benchmark. This paper reports more recent results on a widely used numerical test and on a close-loop configured three-stage hot rolling mill process to reveal the performance of the algorithms at the extreme faulty conditions, and demonstrates the performance under the plant-wide performance supervised framework.

Support Vector Data Description (SVDD) has a limitation for dealing with a large dataset or online learning tasks. This work investigates the practice of credit scoring and proposes a new incremental learning algorithm for SVDD based on Karush-Kuhn-Tucker (KKT) conditions and convex hull. Convex hull and part of newly added samples which violates KKT conditions are treated as new training samples instead of previous support vector and entire new arrived samples. The proposed method can achieve comparable training time with traditional incremental learning algorithm for SVDD while have similar classification accuracy with original SVDD.

In this paper, we firstly study the partial least squares (PLS) and locally weighted project regression (LWPR) methods. As a result, we find out that both methods leave some space for improvement. Then we put forward a new method, called LWPR-MPLS, to improve the traditional statistics and their thresholds. Our method reduces the computation load significantly and therefore can deal with nonlinear problems online. Besides, decomposition of the data in the space without orthogonal residuals brings additional benefits. Simulations show that our method improves the accuracy of fault diagnosis and the capacity of fault classification significantly in comparison with LWPR.

An output-only method for structural diagnosis based on passing vehicle response is proposed in this article. Theoretical derivation reveals that the area of displacement response from only one sensor when vehicle loads pass over a bridge is in relation to global bending stiffness. Therefore, it can be preliminarily ascertained whether damage occurs in structure by comparison of the area of displacement responses before and after damage. Then, in order to localize damage, a damage index composed of displacement responses from several sensors is presented. The damage index has corresponding relationship with local bending stiffness and it is irrelevant to vehicle load parameters. Meanwhile, the method is suitable for both dynamic moving load and static vehicular load test. Case study of a box-girder bridge and experimental data verify the accuracy of the method, and different kinds of vehicle loads at different speeds basically have no influence on the method. In addition, the comparison between the method of curvature of displacement influence line and the presented method is also made, and the results show that the presented damage index is free from the effect of vehicle speeds while the damage index of curvature of displacement influence line is susceptible to vehicle speed.