Training Damage Classifiers in the Absence of Damage Data

机译：在没有损坏数据的情况下训练伤害分类器

获取原文

页面导航

摘要
著录项
相似文献
相关主题

摘要

The main problem associated with Pattern Recognition approaches to Structural Health Monitoring (SHM) is that the higher levels of damage identification almost always require supervised learning. This means that data representing all possible damage states must be available for training purposes, In the case of high-value engineering structures like aircraft, it is simply not possible to generate training data by experiment as it would be prohibitively expensive to damage even one aircraft for the purposes of acquiring data. It is also unlikely that data can always be generated by simulation as the models required would often need to be of such high fidelity that the costs of development and the run-times would again be prohibitive. Recent work by the authors has looked at classifiers which offer the possibility of working robustly on data generated by low-fidelity models or from a simple experimental strategy. The object of this paper is to define the experimental strategy, which simply involves adding masses to the structure and to illustrate its use in feature selection for novelty detection. Despite the fact that the novelty detection approach only requires data from the undamaged structure, the feature selection is facilitated by the availability of damage data. The approach is illustrated using data from an FE simulation.

机译：与模式识别有关的主要问题接近结构健康监测（SHM）是一种更高层次的损伤识别几乎总是需要监督学习。这意味着，代表所有可能损害国家数据必须是可用于训练目的，在高价值的工程结构，如飞机的情况下，根本不可能产生通过实验训练数据，因为这将是非常昂贵的损害，甚至一架飞机用于获取数据的目的。这也是不可能的，数据可以随时通过模拟生成所需的车型，往往就需要是这样高保真的，发展的成本和运行时间将再次让人望而却步。由作者最近的工作看着它提供由低精度模型或从一个简单的实验策略产生的数据强劲工作的可能性分类。本文的目的是定义实验策略，它仅仅包括添加群众的结构和来说明其在特征选择使用新颖性的检测。尽管新颖性检测方法仅需要来自未受损的结构数据时，特征选择被损坏的数据的可用性促进。该方法是使用从FE模拟数据示出。

著录项

来源
《IMAC Conference on Structural Dynamics》|2008年||共12页
会议地点
作者
E. PAPATHEOU; G. MANSON; K. WORDEN;
展开▼
作者单位

展开▼
会议组织
原文格式 PDF
正文语种
中图分类 TH11-53;
关键词

相似文献

外文文献
中文文献
专利

1. Active learning algorithm can establish classifier of blueberry damage with very small training dataset using hyperspectral transmittance data [J] . Hu Meng-Han, Zhao Yu, Zhai Guang-Tao Chemometrics and Intelligent Laboratory Systems . 2018,第期

机译：主动学习算法可以使用超光线透射率数据与非常小的训练数据集建立蓝莓损坏的分类器
2. Experimental chronic low-frequency resistance training produces skeletal muscle hypertrophy in the absence of muscle damage and metabolic stress markers. [J] . Zanchi NE, Lira FS, Seelaender M, Cell biochemistry and function . 2010,第3期

机译：在缺乏肌肉损伤和代谢应激指标的情况下，实验性的慢性低频抵抗训练会导致骨骼肌肥大。
3. Experimental chronic low-frequency resistance training produces skeletal muscle hypertrophy in the absence of muscle damage and metabolic stress markers [J] . Nelo Eidy Zanchi, Fabio Santos Lira, Marilia Seelaender, Cell Biochemistry and Function . 2010,第3期

机译：在缺乏肌肉损伤和代谢应激标志物的情况下，实验性的慢性低频抵抗训练会导致骨骼肌肥大
4. Training Damage Classifiers in the Absence of Damage Data [C] . E. PAPATHEOU, G. MANSON, K. WORDEN IMAC Conference on Structural Dynamics . 2008

机译：在没有损坏数据的情况下训练伤害分类器
5. Absence of large grazers: An obstacle to recovery in acid- and metal-damaged lakes. [D] . Heneberry, Jocelyne H. 1997

机译：缺乏大型放牧者：在酸和金属损坏的湖泊中恢复的障碍。
6. Colocalization of Sensors Is Sufficient to Activate the DNA Damage Checkpoint in the Absence of Damage [O] . Carla Yaneth Bonilla, Justine Amy Melo, David Paul Toczyski -1

机译：传感器的共定位在没有损坏的情况下足以激活DNA损坏检查点
7. Measuring Damage Control Assistant's (DCA) decision-making proficiency in Integrated Damage Control Training Technology (IDCTT) training scenarios. [O] . Fuller John V. 1993

机译：在综合损害控制培训技术（IDCTT）培训方案中衡量损害控制助理（DCA）的决策能力。

Training Damage Classifiers in the Absence of Damage Data

摘要

著录项

相似文献

相关主题

期刊订阅