Method of condition monitoring

• US 7,275,018 B2
• Filed: 08/30/2006
• Issued: 09/25/2007
• Est. Priority Date: 02/10/2006
• Status: Active Grant

• Current Assignee: General Electric Technology GmbH (GE Aerospace)
• Original Assignee: Alstom Technology Ltd. (Alstom SA)
• Inventors: Abu-El-Zeet, Ziad Hasan, Patel, Vijaykumar Chhaganlal
• Primary Examiner(s): Barlow; John
• Assistant Examiner(s): Le; John H.
• Application Number:
  US11/468,332
• Publication Number:
  US 20070192063A1
• Time in Patent Office: 391 Days
• Field of Search: 702/179, 702182-186, 702/188, 702/189, 709/224, 244/100.R, 707/104.1
• US Class Current: 702/182
• CPC Class Codes: G05B 17/02 electric
• CPC Class Codes: G05B 23/0254 based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
• CPC Class Codes: G06Q 10/06 Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
• CPC Class Codes: G06Q 50/06 Energy or water supply

Abstract

A method for the condition monitoring of a fleet of machines, plants, or processes includes the generation of a single generic network for the fleet using historic data from the members of the fleet and using novelty detection tools for analyzing and reducing the data. Scaling factors relevant to each individual fleet member are stored in a scaling file. The fleet members are monitored during operation by comparing test data to the network and calculating quantization errors. Signals on the condition of the fleet members are given based on the quantization errors. The network is adapted by adding new data representing changes in a plant such as due to degradations of components. Due to the individual scaling of data from the different fleet members, the method achieves a sensitivity of monitoring that is comparable to that of methods of the state of the art.

Claims

1. A method of monitoring the condition of a fleet of plants, machines, or processes, the method comprising:

generating and storing a single generic model for all members of the fleet using selected historic data from the fleet taken during normal operation and using novelty detection;

scaling the selected historic data from the individual fleet members by using scaling factors derived from the historic data prior to including them in the single generic model;

monitoring the fleet members during operation by taking test data, scaling the test data using said scaling factors, and comparing the scaled data to the generic model; and

generating a signal on the condition of the fleet members based on the comparison of the measurement data to the generic model;

wherein scaling the input historic data includes subtracting from each data point a mean value and dividing by the standard deviation;

forming clusters of the scaled data;

determining a cluster center for each cluster; and

calculating a cluster threshold for each cluster.

2. The method according to claim 1, wherein forming clusters comprises k-means clustering.

3. The method according to claim 1, further comprising:

storing the mean values and standard deviations as scaling factors in a scaling file for further use in the scaling of test data from each individual fleet member during monitoring.

4. The method according to claim 1, further comprising:

comparing the measured and scaled data to the generic model during monitoring, including establishing a measure of fit using a distance function and determining a quantization error for each measurement data point.

5. The method according to claim 4, wherein establishing a measure of fit comprises establishing with an Euclidian distance function.

6. The method according to claim 4, further comprising:

generating a signal or alert on the condition of a fleet member based on the quantization error.

7. The method according to claim 1, further comprising:

pre-processing the data taken from the fleet members for generating the model, for monitoring the fleet, or both, including removing from the historic data transient data and outliers.

8. A method of monitoring the condition of a fleet of plants, machines, or processes, the method comprising:

generating and storing a single generic model for all members of the fleet using selected historic data from the fleet taken during normal operation and using novelty detection;

scaling the selected historic data from the individual fleet members by using scaling factors derived from the historic data prior to including them in the single generic model;

monitoring the fleet members during operation by taking test data, scaling the test data using said scaling factors, and comparing the scaled data to the generic model;

generating a signal on the condition of the fleet members based on the comparison of the measurement data to the generic model;

adapting the generic model including taking new data;

standardizing the new data by subtracting from each data point a mean value and dividing by a standard deviation to generate new, standardized data;

adding the new, standardized data to the generic model while maintaining all previous data; and

adapting the scaling factors in the scaling file according to the mean values and standard deviations of the new data.

9. The method according to claim 8, further comprising:

determining a number of clusters to be formed of the selected historic data by using a Davies-Bouldin cluster validity algorithm or a visual method that analyzes a plot of sum squared error vs. number of clusters.