• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Society for Machinery Failure Prevention Technology Meeting; 20050418-21; Virginia Beach,VA(US) >A KERNEL BASED REPRESENTATION AND DISCRIMINATION FEATURE EXTRACTION FOR GAS TURBINE FAULT DETECTION AND DIAGNOSIS
【24h】

A KERNEL BASED REPRESENTATION AND DISCRIMINATION FEATURE EXTRACTION FOR GAS TURBINE FAULT DETECTION AND DIAGNOSIS

机译：基于核的代表值和判别特征提取方法在燃气轮机故障检测与诊断中的应用

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Accurate gas turbine fault detection and diagnosis (FDD) are essential for improving airline safety, reducing airline costs associated with delays and cancellations, and scheduling predictive maintenance. We present the maximum representation discrimination features (MRDF) algorithm, a kernel-based nonlinear method of FDD that uses features for maximum representation and discrimination. Examples for training from other classes may not be available or are sparse and one must rely on representation features. When adequate examples for all classes are available, the method will use discriminatory as well as representation features for classification. By varying a single parameter one can judiciously decide the extent to which the features will be used. This method improves upon the existing MRDF classifier by introducing a different between-class distance metric. This metric considers the local neighborhood of points rather than all pairs of points between two classes. This method does not require a nonlinear optimization procedure for its solution; only eigen-decomposition of the kernel matrix is required to arrive at the solution. During operation, the KMRDF classifier will classify new sensor data and make decisions about whether the data indicate presence of fault or not. The developed method was tested with snapshot data, both normal and faulty, collected at takeoff from a turbofan propulsion gas turbine and the results are presented.
机译：准确的燃气轮机故障检测和诊断（FDD）对于提高航空公司安全性，降低与延误和取消相关的航空公司成本以及安排预测性维护至关重要。我们提出了最大表示鉴别特征（MRDF）算法，这是一种基于内核的FDD非线性方法，该方法使用最大表示和鉴别特征。其他课程的培训示例可能不可用或稀疏，并且必须依靠表示功能。当所有类都有足够的示例可用时，该方法将使用区分性和表示性特征进行分类。通过更改单个参数，可以明智地决定使用这些功能的程度。通过引入不同的类间距离度量，此方法对现有MRDF分类器进行了改进。此度量标准考虑的是点的局部邻域，而不是两个类之间的所有成对的点。该方法不需要非线性优化程序即可求解。仅需要核矩阵的特征分解即可得出解。在操作过程中，KMRDF分类器将对新的传感器数据进行分类，并对数据是否表明存在故障做出​​决策。对开发的方法进行了测试，并从涡轮风扇推进燃气轮机起飞时收集了正常和故障的快照数据，并给出了结果。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号