Generalized autoencoder-based fault detection method for traction systems with performance degradation

Fault diagnosis of traction systems is important for the safety operation of high-speed trains.Long-term operation of the trains will degrade the performance of systems,which decreases the fault detection accuracy.To solve this problem,this paper proposes a fault detection method developed by a Generalized Autoencoder(GAE)for systems with performance degradation.The advantage of this method is that it can accurately detect faults when the traction system of high-speed trains is affected by performance degradation.Regardless of the probability distribution,it can handle any data,and the GAE has extremely high sensitivity in anomaly detection.Finally,the effectiveness of this method is verified through the Traction Drive Control System(TDCS)platform.At different performance degradation levels,our method’s experimental results are superior to traditional methods.

A Study of Ensemble Feature Selection and Adversarial Training for Malicious User Detection

The continuously booming of information technology has shed light on developing a variety of communication networks,multimedia,social networks and Internet of Things applications.However,users inevitably suffer from the intrusion of malicious users.Some studies focus on static characteristics of malicious users,which is easy to be bypassed by camouflaged malicious users.In this paper,we present a malicious user detection method based on ensemble feature selection and adversarial training.Firstly,the feature selection alleviates the dimension disaster problem and achieves more accurate classification performance.Secondly,we embed features into the multidimensional space and aggregate it into a feature map to encode the explicit content preference and implicit interaction preference.Thirdly,we use an effective ensemble learning which could avoid over-fitting and has good noise resistance.Finally,we propose a datadriven neural network detection model with the regularization technique adversarial training to deeply analyze the characteristics.It simplifies the parameters,obtaining more robust interaction features and pattern features.We demonstrate the effectiveness of our approach with numerical simulation results for malicious user detection,where the robustness issues are notable concerns.

Deep Domain-Adversarial Anomaly Detection With One-Class Transfer Learning

Despite the big success of transfer learning techniques in anomaly detection,it is still challenging to achieve good transition of detection rules merely based on the preferred data in the anomaly detection with one-class classification,especially for the data with a large distribution difference.To address this challenge,a novel deep one-class transfer learning algorithm with domain-adversarial training is proposed in this paper.First,by integrating a hypersphere adaptation constraint into domainadversarial neural network,a new hypersphere adversarial training mechanism is designed.Second,an alternative optimization method is derived to seek the optimal network parameters while pushing the hyperspheres built in the source domain and target domain to be as identical as possible.Through transferring oneclass detection rule in the adaptive extraction of domain-invariant feature representation,the end-to-end anomaly detection with one-class classification is then enhanced.Furthermore,a theoretical analysis about the model reliability,as well as the strategy of avoiding invalid and negative transfer,is provided.Experiments are conducted on two typical anomaly detection problems,i.e.,image recognition detection and online early fault detection of rolling bearings.The results demonstrate that the proposed algorithm outperforms the state-of-the-art methods in terms of detection accuracy and robustness.

Unsupervised Anomaly Detection Approach Based on Adversarial Memory Autoencoders for Multivariate Time Series

The widespread usage of Cyber Physical Systems(CPSs)generates a vast volume of time series data,and precisely determining anomalies in the data is critical for practical production.Autoencoder is the mainstream method for time series anomaly detection,and the anomaly is judged by reconstruction error.However,due to the strong generalization ability of neural networks,some abnormal samples close to normal samples may be judged as normal,which fails to detect the abnormality.In addition,the dataset rarely provides sufficient anomaly labels.This research proposes an unsupervised anomaly detection approach based on adversarial memory autoencoders for multivariate time series to solve the above problem.Firstly,an encoder encodes the input data into low-dimensional space to acquire a feature vector.Then,a memory module is used to learn the feature vector’s prototype patterns and update the feature vectors.The updating process allows partial forgetting of information to prevent model overgeneralization.After that,two decoders reconstruct the input data.Finally,this research uses the Peak Over Threshold(POT)method to calculate the threshold to determine anomalous samples from normal samples.This research uses a two-stage adversarial training strategy during model training to enlarge the gap between the reconstruction error of normal and abnormal samples.The proposed method achieves significant anomaly detection results on synthetic and real datasets from power systems,water treatment plants,and computer clusters.The F1 score reached an average of 0.9196 on the five datasets,which is 0.0769 higher than the best baseline method.

Colonoscopy procedural volume increases adenoma and polyp detection rates in gastroenterology trainees

AIM To investigate changes in polyp detection throughout fellowship training, and estimate colonoscopy volume required to achieve the adenoma detection rate(ADRs) and polyp detection rate(PDRs) of attending gastroenterologists.METHODS We reviewed colonoscopies from July 1, 2009 to June 30, 2014. Fellows' procedural logs were used to retrieve colonoscopy procedural volumes, and these were treated as the time variable. Findings from screening colonoscopies were used to calculate colonoscopy outcomes for each fellow for the prior 50 colonoscopies at each time point. ADR and PDR were plotted against colonoscopy procedural volumes to produce individual longitudinal graphs. Repeated measures linear mixed effects models were used to study the change of ADR and PDR with increasing procedural volume.RESULTS During the study period, 12 fellows completed full three years of training and were included in the analysis. The average ADR and PDR were, respectively, 31.5% and 41.9% for all fellows, and 28.9% and 38.2% for attendings alone. There was a statistically significant increase in ADR with increasing procedural volume(1.8%/100 colonoscopies, P = 0.002). Similarly, PDR increased 2.8%/100 colonoscopies(P = 0.0001), while there was no significant change in advanced ADR(0.04%/100 colonoscopies, P = 0.92). The ADR increase was limited to the right side of the colon, while the PDR increased in both the right and left colon. The adenoma per colon and polyp per colon also increased throughout training. Fellows reached the attendings' ADR and PDR after 265 and 292 colonoscopies, respectively.CONCLUSION We found that the ADR and PDR increase with increasing colonoscopy volume throughout fellowship. Our findings support recent recommendations of ≥ 275 colonoscopies for colonoscopy credentialing.

Adversarial Training Against Adversarial Attacks for Machine Learning-Based Intrusion Detection Systems

Intrusion detection system plays an important role in defending networks from security breaches.End-to-end machine learning-based intrusion detection systems are being used to achieve high detection accuracy.However,in case of adversarial attacks,that cause misclassification by introducing imperceptible perturbation on input samples,performance of machine learning-based intrusion detection systems is greatly affected.Though such problems have widely been discussed in image processing domain,very few studies have investigated network intrusion detection systems and proposed corresponding defence.In this paper,we attempt to fill this gap by using adversarial attacks on standard intrusion detection datasets and then using adversarial samples to train various machine learning algorithms(adversarial training)to test their defence performance.This is achieved by first creating adversarial sample based on Jacobian-based Saliency Map Attack(JSMA)and Fast Gradient Sign Attack(FGSM)using NSLKDD,UNSW-NB15 and CICIDS17 datasets.The study then trains and tests JSMA and FGSM based adversarial examples in seen(where model has been trained on adversarial samples)and unseen(where model is unaware of adversarial packets)attacks.The experiments includes multiple machine learning classifiers to evaluate their performance against adversarial attacks.The performance parameters include Accuracy,F1-Score and Area under the receiver operating characteristic curve(AUC)Score.

Monitoring energy usage of heavy-haul iron ore trains with on-board energy meter for improving energy efficiency

Purpose-For billing purposes,heavy-haul locomotives in Sweden are equipped with on-board energy meters,which can record several parameters,e.g.,used energy,regenerated energy,speed and position.Since there is a strong demand for improving energy efficiency in Sweden,data from the energy meters can be used to obtain a better understanding of the detailed energy usage of heavy-haul trains and identify potential for future improvements.Design/methodology/approach-To monitor energy efficiency,the present study,therefore,develops key performance indicators(KPIs),which can be calculated with energy meter data to reflect the energy efficiency of heavy-haul trains in operation.Energy meter data of IORE class locomotives,hauling highly uniform 30-tonne axle load trains with 68 wagons,together with additional data sources,are analysed to identify significant parameters for describing driver influence on energy usage.Findings-Results show that driver behaviour varies significantly and has the single largest influence on energy usage.Furthermore,parametric studies are performed with help of simulation to identify the influence of different operational and rolling stock conditions,e.g.,axle loads and number of wagons,on energy usage.Originality/value-Based on the parametric studies,some operational parameters which have significant impact on energy efficiency are found and then the KPIs are derived.In the end,some possible measures for improving energy performance in heavy-haul operations are given.

Stream-computing of High Accuracy On-board Real-time Cloud Detection for High Resolution Optical Satellite Imagery

This paper focuses on the time efficiency for machine vision and intelligent photogrammetry, especially high accuracy on-board real-time cloud detection method. With the development of technology, the data acquisition ability is growing continuously and the volume of raw data is increasing explosively. Meanwhile, because of the higher requirement of data accuracy, the computation load is also becoming heavier. This situation makes time efficiency extremely important. Moreover, the cloud cover rate of optical satellite imagery is up to approximately 50%, which is seriously restricting the applications of on-board intelligent photogrammetry services. To meet the on-board cloud detection requirements and offer valid input data to subsequent processing, this paper presents a stream-computing of high accuracy on-board real-time cloud detection solution which follows the “bottom-up” understanding strategy of machine vision and uses multiple embedded GPU with significant potential to be applied on-board. Without external memory, the data parallel pipeline system based on multiple processing modules of this solution could afford the “stream-in, processing, stream-out” real-time stream computing. In experiments, images of GF-2 satellite are used to validate the accuracy and performance of this approach, and the experimental results show that this solution could not only bring up cloud detection accuracy, but also match the on-board real-time processing requirements.

Association of trainee participation with adenoma and polyp detection rates

To investigate whether adenoma and polyp detection rates (ADR and PDR, respectively) in screening colonoscopies performed in the presence of fellows differ from those performed by attending physicians alone. METHODSWe performed a retrospective review of all patients who underwent a screening colonoscopy at Grady Memorial Hospital between July 1, 2009 and June 30, 2015. Patients with a history of colon polyps or cancer and those with poor colon preparation or failed cecal intubation were excluded from the analysis. Associations of fellowship training level with the ADR and PDR relative to attendings alone were assessed using unconditional multivariable logistic regression. Models were adjusted for sex, age, race, and colon preparation quality. RESULTSA total of 7503 colonoscopies met the inclusion criteria and were included in the analysis. The mean age of the study patients was 58.2 years; 63.1% were women and 88.2% were African American. The ADR was higher in the fellow participation group overall compared to that in the attending group: 34.5% vs 30.7% (P = 0.001), and for third year fellows it was 35.4% vs 30.7% (aOR = 1.23, 95%CI: 1.09-1.39). The higher ADR in the fellow participation group was evident for both the right and left side of the colon. For the PDR the corresponding figures were 44.5% vs 40.1% (P = 0.0003) and 45.7% vs 40.1% (aOR = 1.25, 95%CI: 1.12-1.41). The ADR and PDR increased with increasing fellow training level (P for trend < 0.05). CONCLUSIONThere is a stepwise increase in ADR and PDR across the years of gastroenterology training. Fellow participation is associated with higher adenoma and polyp detection.

Detection of butt weld of laser-MIG hybrid welding of thin-walled profile for high-speed train

Purpose–This study aims to solve the problem of weld quality inspection,for the aluminum alloy profile welding structure of high-speed train body has complex internal shape and thin plate thickness(2–4 mm),the conventional nondestructive testing method of weld quality is difficult to implement.Design/methodology/approach–In order to solve this problem,the ultrasonic creeping wave detection technology was proposed.The impact of the profile structure on the creeping wave detection was studied by designing profile structural test blocks and artificial simulation defect test blocks.The detection technology was used to test the actual welded test blocks,and compared with the results of X-ray test and destructive test(tensile test)to verify the accuracy of the ultrasonic creeping wave test results.Findings–It is indicated that that X-ray has better effect on the inspection of porosities and incomplete penetration defects.However,due to special detection method and protection,the detection speed is slow,which cannot meet the requirements of field inspection of the welding structure of aluminum alloy thin-walled profile for high-speed train body.It can be used as an auxiliary detection method for a small number of sampling inspection.The ultrasonic creeping wave can be used to detect the incomplete penetration welds with the equivalent of 0.25 mm or more,the results of creeping wave detection correspond well with the actual incomplete penetration defects.Originality/value–The results show that creeping wave detection results correspond well with the actual non-penetration defects and can be used for welding quality inspection of aluminum alloy thin-wall profile composite welding joints.It is recommended to use the echo amplitude of the 10 mm 30.2 mm 30.5 mm notch as the criterion for weld qualification.

An Approach to Detect Structural Development Defects in Object-Oriented Programs

Structural development defects essentially refer to code structure that violates object-oriented design principles. They make program maintenance challenging and deteriorate software quality over time. Various detection approaches, ranging from traditional heuristic algorithms to machine learning methods, are used to identify these defects. Ensemble learning methods have strengthened the detection of these defects. However, existing approaches do not simultaneously exploit the capabilities of extracting relevant features from pre-trained models and the performance of neural networks for the classification task. Therefore, our goal has been to design a model that combines a pre-trained model to extract relevant features from code excerpts through transfer learning and a bagging method with a base estimator, a dense neural network, for defect classification. To achieve this, we composed multiple samples of the same size with replacements from the imbalanced dataset MLCQ1. For all the samples, we used the CodeT5-small variant to extract features and trained a bagging method with the neural network Roberta Classification Head to classify defects based on these features. We then compared this model to RandomForest, one of the ensemble methods that yields good results. Our experiments showed that the number of base estimators to use for bagging depends on the defect to be detected. Next, we observed that it was not necessary to use a data balancing technique with our model when the imbalance rate was 23%. Finally, for blob detection, RandomForest had a median MCC value of 0.36 compared to 0.12 for our method. However, our method was predominant in Long Method detection with a median MCC value of 0.53 compared to 0.42 for RandomForest. These results suggest that the performance of ensemble methods in detecting structural development defects is dependent on specific defects.

Face Detection Detection, Alignment Alignment, Quality Assessment and Attribute Analysis with Multi-Task Hybrid Convolutional Neural Networks

This paper proposes a universal framework,termed as Multi-Task Hybrid Convolutional Neural Network(MHCNN),for joint face detection,facial landmark detection,facial quality,and facial attribute analysis.MHCNN consists of a high-accuracy single stage detector(SSD)and an efficient tiny convolutional neural network(T-CNN)for joint face detection refinement,alignment and attribute analysis.Though the SSD face detectors achieve promising results,we find that applying a tiny CNN on detections further boosts the detected face scores and bounding boxes.By multi-task training,our T-CNN aims to provide five facial landmarks,facial quality scores,and facial attributes like wearing sunglasses and wearing masks.Since there is no public facial quality data and facial attribute data as we need,we contribute two datasets,namely FaceQ and FaceA,which are collected from the Internet.Experiments show that our MHCNN achieves face detection performance comparable to the state of the art in face detection data set and benchmark(FDDB),and gets reasonable results on AFLW,FaceQ and FaceA.

Algorithm of Helmet Wearing Detection Based on AT-YOLO Deep Mode

The existing safety helmet detection methods are mainly based on one-stage object detection algorithms with high detection speed to reach the real-time detection requirements,but they can’t accurately detect small objects and objects with obstructions.Therefore,we propose a helmet detection algorithm based on the attention mechanism(AT-YOLO).First of all,a channel attention module is added to the YOLOv3 backbone network,which can adaptively calibrate the channel features of the direction to improve the feature utilization,and a spatial attention module is added to the neck of the YOLOv3 network to capture the correlation between any positions in the feature map so that to increase the receptive field of the network.Secondly,we use DIoU(Distance Intersection over Union)bounding box regression loss function,it not only improving the measurement of bounding box regression loss but also increases the normalized distance loss between the prediction boxes and the target boxes,which makes the network more accurate in detecting small objects and faster in convergence.Finally,we explore the training strategy of the network model,which improves network performance without increasing the inference cost.Experiments show that the mAP of the proposed method reaches 96.5%,and the detection speed can reach 27 fps.Compared with other existing methods,it has better performance in detection accuracy and speed.

Virtual simulation experiment of the design and manufacture of a beer bottle-defect detection system

Background Machine learning-based beer bottle-defect detection is a complex technology that runs automatically;however,it consumes considerable memory,is expensive,and poses a certain danger when training novice operators.Moreover,some topics are difficult to learn from experimental lectures,such as digital image processing and computer vision.However,virtual simulation experiments have been widely used to good effect within education.A virtual simulation of the design and manufacture of a beer bottle-defect detection system will not only help the students to increase their image-processing knowledge,but also improve their ability to solve complex engineering problems and design complex systems.Methods The hardware models for the experiment(camera,light source,conveyor belt,power supply,manipulator,and computer)were built using the 3DS MAX modeling and animation software.The Unreal Engine 4(UE4)game engine was utilized to build a virtual design room,design the interactive operations,and simulate the system operation.Results The results showed that the virtual-simulation system received much better experimental feedback,which facilitated the design and manufacture of a beer bottle-defect detection system.The specialized functions of the functional modules in the detection system,including a basic experimental operation menu,power switch,image shooting,image processing,and manipulator grasping,allowed students(or virtual designers)to easily build a detection system by retrieving basic models from the model library,and creating the beer-bottle transportation,image shooting,image processing,defect detection,and defective-product removal.The virtual simulation experiment was completed with image processing as the main body.Conclusions By mainly focusing on bottle mouth defect detection,the detection system dedicates more attention to the user and the task.With more detailed tasks available,the virtual system will eventually yield much better results as a training tool for image processing education.In addition,a novel visual perception-thinking pedagogical framework enables better comprehension than the traditional lecture-tutorial style.

Fall detection system in enclosed environments based on single Gaussian model

In this paper,we propose an efficient fall detection system in enclosed environments based on single Gaussian model using the maximum likelihood method.Online video clips are used to extract the features from two cameras.After the model is constructed,a threshold is set,and the probability for an incoming sample under the single Gaussian model is compared with that threshold to make a decision.Experimental results show that if a proper threshold is set,a good recognition rate for fall activities can be achieved.

基于改进YOLOv8的地铁列车焊缝缺陷轻量化检测方法

针对现有的地铁列车车体焊接质量检测技术存在检测模型较大、检测精度和效率较低的问题,提出一种基于改进YOLOv8的焊缝缺陷轻量化检测方法。首先,利用相控阵超声波检测仪采集对接焊缝内部缺陷图像,通过图像预处理制作成焊缝缺陷数据集。然后,在YOLOv8模型基础上,利用Inner-SIoU优化原有损失函数、采用C2f-PConv替换C2f模块、引入大型可分离核注意力(LSKA)模块和挤压激励(SE)注意力机制,建立了基于改进YOLOv8的地铁列车车体焊缝缺陷质量检测模型,以提高焊缝缺陷特征提取和多尺度特征融合的能力。最后,利用改进的YOLOv8模型对焊缝缺陷数据集进行训练和测试。结果表明,改进的YOLOv8模型大小为7.91 M,对于焊缝缺陷的检测精度达到98.30%,检测速度达到138.9帧/s,与YOLOv8原始模型相比,模型更小,检测精度更高。

贫数据中基于模型自训练的空气处理设备故障诊断

针对空气处理设备(AHU)故障贫数据,基于深度置信网络(DBN)模型对4种特征选择算法进行对比研究,结果表明最大相关最小冗余算法的特征子集在诊断准确率及子集元素稳定性上表现最优。提出将DBN嵌入自训练框架的故障诊断模型,发现DBN自训练的诊断准确率较单纯DBN最高可提升19.5%。提出均匀抽样及按比例抽样2种自训练伪标签抽样策略,二者的诊断准确率均随抽样数减小而增大,在不同抽样数中的最大差异为3.42%;在所有贫数据样本中,均匀抽样策略始终优于按比例抽样,诊断准确率最大相差1.39%,表明在故障标签匮乏时,采用均匀抽样策略及较小的抽样数有利于提升DBN自训练的诊断性能。

基于生成对抗Transformer的电力负荷数据异常检测

电力负荷异常数据将给电力系统规划、负荷预测以及用能分析等带来较大的负面影响,因此亟须对负荷数据异常进行检测与识别。首先,针对电力负荷数据异常分类、原因及其特征开展分析。其次,改进传统Transformer编码器结构,采用多头注意力层代替掩码多头注意力层,同时移除前馈网络,以提高模型对负荷时序序列的全局注意力。基于生成对抗网络(generative adversarial networks,GAN)生成器与判别器的博弈结构,提出一种改进的GAN-Transformer模型,以更好地捕捉趋势性特征并加速模型收敛。然后,引入多阶段映射与训练方法,综合焦点分数打分机制,通过分阶段负荷序列重构帮助模型更好地提取负荷数据异常特征。最后,算例分析结果表明,GAN-Transformer模型在负荷数据异常检测精确率、召回率、F_(1)值以及训练时间方面均具有更优的性能,验证了所提方法的有效性和优越性。文中研究工作为基于深度学习进一步实现电力负荷数据异常分类与数据修复提供了有益参考。

课程学习指导下的半监督目标检测框架

为了提高伪标签的质量,解决半监督目标检测(SSOD)中的确认偏差问题,并针对现有算法中忽视无标注数据复杂性导致错误伪标签的难点,提出一种课程学习(CL)指导下的SSOD框架,该框架主要由ICSD(IoUConfidence-Standard-Deviation)难度测量器和BP(Batch-Package)训练调度器这2个模块组成。其中,ICSD难度测量器综合考虑了伪边界框之间的交并比(IoU)、置信度、类别标签等信息,并引入C_IOU(Checkpoint_IOU)方法评估无标注数据的可靠性;BP训练调度器设计2种高效调度策略,分别从Batch和Package角度出发,优先选择可靠性指标高的无标记数据,实现以CL的方式充分利用整个无标记数据集。在Pascal VOC和MS-COCO数据集上的广泛对比实验结果表明,所提框架不仅适用于现有的SSOD算法,而且检测精度和稳定性都得到显著提升。