Anomaly Detection for Prognostic and Health Management .ppt
《Anomaly Detection for Prognostic and Health Management .ppt》由会员分享,可在线阅读,更多相关《Anomaly Detection for Prognostic and Health Management .ppt(35页珍藏版)》请在麦多课文档分享上搜索。
1、Anomaly Detection for Prognostic and Health Management System Development,Tom Brotherton,New Stealth Technology,Outline,What is Anomaly Detection Different types of anomaly detectorsRadial Basis Function Neural Net Anomaly Detector The basics Comparison with other neural net approaches Feature off-n
2、ominal distance measures TrainingImplementations Continuous = Gas turbine engine monitoring Snap shot = Web server helicopter vibration condition indicators RBF NN & Boxplots Application to detection of helicopter bearing fault Application to monitoring fish behavior for water quality monitoring,Wha
3、t is Anomaly Detection?,Anomaly Detection = The Detection of Any Off-Nominal Event Data Known fault conditions Novel event = New - never seen before data New type of fault New variation of known nominal or fault data What is Nominal Sets of parameters that behave as expected Physics models Statistic
4、al models,Approaches,Applicability,Physics,Parametric - Estimate of physics,Empirical - Derived from collected data,State Variable Models (derived from physics),JPL: BEAM (coherence = model of linear relationships),Neural nets (non-linear relationships),Academic: Support Vector,Ex: Gas Turbine Engin
5、e Deck: Component level physics model,Simple statistics,Hybrid Model: Combine Physics + Empirical,Fused empirical: BEAM + NN,Empirical Modeling,Collected Nominal Data,Idea: Theoretical boundary (multi-dimensional tube) that data should lie within: - Nominal data is inside the boundary - Anomaly data
6、 is outside,Problem: How to estimate / approximate the boundary?,An anomaly,Problem: What measurement(s) caused the anomaly?,Problem: How far off-nominal is the anomaly / feature?,RBF Neural Net Anomaly Detection: The Idea,Dynamic data = Lots of NN basis units to model Piecewise stationary approxima
7、tion Distance measure = Function of the signal set Individual signal distances from nominal = distance from “closest” basis unit Detection can be for set of signals when no single signal is anomalous The model can be adaptively updated to include additional data / known fault classes Trajectories of
8、 features relative to basis unit = Prognosis,Radial Basis Function (RBF) Neural Net Model,Why Use Radial Basis Function Neural Nets?,Radial Basis Function Neural Net Nearest neighbor classifier Distance metric : Measure “nominal” Multi-layer perceptron (MLP) does not have these properties,Support Ve
9、ctor Machine,In some sense, much better model of truth . but Automated selection of number of basis units Lots! Trade off between fidelity vs smoothness Not practical for on-wing How to compute individual signal distances Loss of intuition,Training data,NN = Model for Nominal Data,Feature Distance C
10、alculation,?,Nearest Neighbor Distance,NN = Model for Nominal Data,Alternative Distance Calculation,Alternative Distance = Which Basis Unit gives the smallest number of individual off-nominal features - Hamming Distance (from digital communications decoding),RBF NN Architectures,Gaussian elliptical
11、basis function :,Fuzzy membership basis function :,Rayleigh basis function :,Detector Output,= Gaussian Mixture Model,Good for magnitude spectral data * Basis function is matched to the data distribution,For those who like things fuzzy,Small number of clusters Small number of basis units Low False A
12、larms, Very general Missed detectionsToo General ?,- Large number of clusters Good tracking of data dynamics Large number of basis units, More sensitive to outliers More false alarmsOver Trained ?,Dont know a-priori what are the best settings,Training : Neural Net Architectures How to select paramet
13、ers,RBF Training,Cluster the data to form Basis Units K-means clustering Assumes no a-priori knowledge of data relationships Optimization to determine centers and included points Alternative Clustering Take advantage of fact that data is continuous in time Clusters will be contiguous in time Determi
14、nistic so no optimization required 500xs faster the K-means cluster Weights are found via LMS estimate,M of N Detection,Detection? False alarm?,Large scale factor,Trade off single point detection capability vs false alarm rateLarge Scale Factor / Small N Short high SNR anomaliesSmall Scale Factor /
15、Large N Long persistent low SNR anomalies,Idea: M of N detection allows one sample high false alarm rate Then integrate over time to remove,Alternatives,This technique works well Demonstrated by Pratt & Whitney for C-17 F117 applications Transient engine operations Long time to train lots of differe
16、nt types of transients Model can become very complex Engine control system On-wing memory and timing constraints Alternative Combine equipment operating regime recognition with anomaly detector Ex: Identify steady operation and then take a snapshot of the data Simple statistics may suffice,Example G
17、as Turbine Operations,Regime recognition Regimes: Transient Throttle up Transient Throttle down Steady state B14 open Steady state B14 closed,Break the big problem in to a set of small problems,Anomaly Detection of Stationary Regime Detected Data,Web Server Implementation for Helicopter Vibration Da
- 1.请仔细阅读文档,确保文档完整性,对于不预览、不比对内容而直接下载带来的问题本站不予受理。
- 2.下载的文档,不会出现我们的网址水印。
- 3、该文档所得收入(下载+内容+预览)归上传者、原创作者;如果您是本文档原作者,请点此认领!既往收益都归您。
下载文档到电脑,查找使用更方便
2000 积分 0人已下载
下载 | 加入VIP,交流精品资源 |
- 配套讲稿:
如PPT文件的首页显示word图标,表示该PPT已包含配套word讲稿。双击word图标可打开word文档。
- 特殊限制:
部分文档作品中含有的国旗、国徽等图片,仅作为作品整体效果示例展示,禁止商用。设计者仅对作品中独创性部分享有著作权。
- 关 键 词:
- ANOMALYDETECTIONFORPROGNOSTICANDHEALTHMANAGEMENTPPT
链接地址:http://www.mydoc123.com/p-378424.html