Research Papers

Growing Structure Multiple Model Systems for Anomaly Detection and Fault Diagnosis

[+] Author and Article Information
Jianbo Liu

Manufacturing Systems Research Laboratory, General Motors Research and Development, 30500 Mound Road, Warren MI 48090jianbo.liu@gm.com

Dragan Djurdjanovic1

Department of Mechanical Engineering, University of Texas, Austin, TX 78712dragand@me.utexas.edu

Kenneth Marko

 ETAS Inc., Ann Arbor, MI 48103ken.marko@etas.us

Jun Ni

Department of Mechanical Engineering, University of Michigan, Ann Arbor, MI 48109junni@umich.edu

The larger the overlap between two distributions is, the more samples are needed to identify with statistical significance the corresponding system condition.


Corresponding author.

J. Dyn. Sys., Meas., Control 131(5), 051001 (Aug 17, 2009) (13 pages) doi:10.1115/1.3155004 History: Received May 03, 2006; Revised January 06, 2009; Published August 17, 2009

A new anomaly detection scheme based on growing structure multiple model system (GSMMS) is proposed in this paper to detect and quantify the effects of anomalies. The GSMMS algorithm combines the advantages of growing self-organizing networks with efficient local model parameter estimation into an integrated framework for modeling and identification of general nonlinear dynamic systems. The identified model then serves as a foundation for building an effective anomaly detection and fault diagnosis system. By utilizing the information about system operation region provided by the GSMMS, the residual errors can be analyzed locally within each operation region. This local decision making scheme can accommodate for unequally distributed residual errors across different operational regions. The performance of the newly proposed method is evaluated through anomaly detection and quantification in an electronically controlled throttle system, which is simulated using a high-fidelity engine simulation software package provided by a major automotive manufacturer for control system development.

Copyright © 2009 by American Society of Mechanical Engineers
Your Session has timed out. Please sign back in to continue.



Grahic Jump Location
Figure 1

Partition of input-output space using self-organizing network; (a) Voronoi tessellation and (b) self-organizing network

Grahic Jump Location
Figure 2

Illustration of the training procedure for the growing structure multiple model system and the breadth-first algorithm for calculating the shortest distance between the best matching unit and its neighboring nodes on the self-organizing network

Grahic Jump Location
Figure 3

Flow chart of the sequential training algorithm for growing structure multiple model system

Grahic Jump Location
Figure 4

Overview of region dependent decision making scheme

Grahic Jump Location
Figure 5

Illustration of anomaly detection and subsequent fault identification

Grahic Jump Location
Figure 6

Electronic throttle plate. Mc and Mp are the model for the controller and plant respectively. ADc and ADp are the corresponding anomaly detectors.

Grahic Jump Location
Figure 7

The 3D plots of self-organizing networks for the plant and the controller: (a) plant and (b) controller

Grahic Jump Location
Figure 8

Output errors during normal system operation: (a) plant and (b) controller

Grahic Jump Location
Figure 9

Output errors during abnormal system operation: (a) 10% reduction in plant gain, (b) 15% reduction in plant stiffness, and (c) 10% reduction in controller gain

Grahic Jump Location
Figure 10

Output error densities in six adjacent regions from controller anomaly detector when controller gain is reduced by 5%, 10%, 15%, and 20%

Grahic Jump Location
Figure 11

The confidence values calculated in six adjacent regions from controller anomaly detector when the controller gain is gradually reduced from its nominal value

Grahic Jump Location
Figure 12

Area under the ROC curves of plant anomaly detector for detecting two types of anomalies. The ROC curves have also been plotted for detecting 10% parameter reductions in both cases. TP rate is the true positive rate and FP rate is the false positive rate. (a) Plant stiffness reduction and (b) plant gain reduction.

Grahic Jump Location
Figure 13

Error densities from three different operation regions of the plant under normal and faulty operations. F1 represents that plant gain has been reduced by 20%. F2 represents that plant stiffness is reduced by 20%.




Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In