Few-shot aerial image object detection aims to rapidly detect instances of novel category in images by using few labeled samples. However, due the complex background images, samples categories, and model trained with few-shot learning paradigm is biased towards base it greatly increases difficulty identifying foreground objects categories. In addition this, tiny always a hot potato detection, even more difficult for detection. To this end, we propose Confidence-Iou collaborative proposal...

In recent years, network-based resilience assessment has aroused attention because of its strong link to the stability and dependability complex systems. Previous studies have contributed definition quantification system resilience, but an integral consistent framework is still lacking for procedure analysis general systems, responses strains induced by multiple rounds disruptions not been well studied. this manuscript, dynamic defined as a system's ability resist loss adapt successive...

Built-in test (BIT) technology provides an effective way of state monitoring, and is widely used in many fields. However, the false alarm prevents its wider usage reduces reliability. The existing BIT methods typically assume that probability generating a stationary process, which far too simplifying assumption. In this article, we propose new method recognition, considering time series characteristics BIT. presented divides evolution process into three phases, recognition intermediate phase...

摘要: 在智能制造背景下,离散制造企业对利用大数据技术提高车间生产管控水平提出了迫切的需求。研究大数据驱动的离散制造车间生产过程智能管控方法,在明确离散制造车间特点与管控需求的基础上,分析了传统方法的局限性和大数据方法的优势,进而提出大数据驱动的离散制造车间生产过程管控总体框架,以制造大数据的"采集-处理-分析-服务"为主线开展研究。在"进度预测-瓶颈发现-异常溯源-智能决策"的生产过程闭环管控机制中,分别提出:基于堆叠稀疏自编码机的生产进度在线预测技术,基于平行门控循环单元的生产瓶颈漂移发现技术,基于密度峰值-模糊C均值的生产异常溯源分析技术和基于多智能体强化学习的生产过程智能决策技术。最后,以某航空企业典型离散制造车间作为对象,对所提出的大数据分析与智能决策方法进行了原型系统开发和应用验证。

Due to the hash working environment of Special Vehicles (SV), intelligent operation is necessary monitor and keep healthy status SV. Digital Twin (DT) a key enabling technology towards for In order support DT construction applications, data treatment process designed in this study, which mainly includes modeling integration. Firstly, overall architecture constructed illustrate supportive relations from collection applications. Secondly, spatio-temporal model presented organize SV spatial,...

Traditional fault detection approaches for aeroengines based on PCA cannot effectively detect faults when the data does not follow normal distribution. Meanwhile, there are few effective methods elimination of outliers during modeling thus model precision be guaranteed. Aiming at a solution problems above, new approach aero-engines is proposed, including PCA-KDE and R-PCA outliers' approach. The instance analysis turbofan engine shows that this able to potential accurately can provide...

The dependency matrix is widely used in fault diagnosis as a valid diagnostic model. However, the conventional matrix-based method pays more attention to extension and improvement of matrix, timeliness rarely discussed. To solve problem improve efficiency, fast using two search algorithms proposed. hash function introduced, transformed into key-value pairs. binary algorithm brought locate desired failure mode by splitting matrix. Thus, converted process, which can efficiency making full use...

Abstract The Non-electric stimulus transfer system (NSTS) is a sensitivity product known for its high reliability. To enhance the accuracy of reliability assessment NSTS, this study investigates relationship between NSTS and initiating devices, proposes synthetic method. By considering working principle failure logic typical function-shared units are introduced to model NSTS. Subsequently, approximated by using normal distribution, incorporating evaluation devices test data derive an...

Knowledge inference and knowledge prediction is widely used in the intelligent fault diagnosis that very important to product safety. Most learning graph embedding methods represent entities relations only with fact triples of graphs (KGs) through translating models without integrating rich semantic information entity descriptions. However, description-based method DKRL takes high frequency words descriptions as input data training using CNN encoder model, which loss word order feature...

A new approach to false alarm recognition is proposed. The described method divides the state of a system into three types: normal, false-alarm, and faulty, analyzes overlapping relations distribution functions different states determine optimal thresholds. After brief introduction support vector machine (SVM), proposed strategy based on results using thresholds explained. presented evolutionary illustrated by fault injection system. accuracy technique compared with conventional other...

Around the "fault diagnosis" to analyze manufacturing system reliability at device performance level, it is increasingly unable meet requirements of predictive production. To achieve refined and integrated process control optimization, a dynamic mission modeling method for systems with multiple production lines proposed in this paper. First, based on operating mechanism its composition, connection covering equipment, material, task proposed. Second, mapping decomposition load network model...

Accurate 3D urban scene reconstruction from a set of multi-view images is an important and challenging task in various fields such as computer visions graphics. The previous methods are usually dilemma when considering the problems completeness accuracy, especially for complex outdoor scenes. To address these issues, this paper we propose new framework based on number available methods. Apart joint optimization with image semantic segmentation, also attempt to introduce unified evaluation...