Machine performance monitoring and fault classification using an exponentially weighted moving average scheme

• US 5,602,761 A
• Filed: 12/30/1993
• Issued: 02/11/1997
• Est. Priority Date: 12/30/1993
• Status: Expired due to Term

• Current Assignee: Caterpillar Incorporated
• Original Assignee: Caterpillar Incorporated
• Inventors: Lin, Chang-Ching, Spoerre, Julie K., Wang, Hsu-Pin
• Primary Examiner(s): Voeltz, Emanuel T.
• Assistant Examiner(s): Kemper, M.
• Application Number:
  US08/176,456
• Time in Patent Office: 1,139 Days
• Field of Search: 364/184, 364/508, 364/581, 364/474.16, 364/141, 364/551.01, 364/551.02, 364/575, 340/679
• US Class Current: 702/179
• CPC Class Codes: G01H 1/003 of rotating machines G01H1/10 takes precedence
• CPC Class Codes: G01M 13/045 Acoustic or vibration analysis

Abstract

The present invention provides an accurate machine monitoring technique based on vibration analysis. An AR parametric model is generated to characterize a normal machine condition. Subsequently, data is collected from a machine during operation. This data is fit to the AR parametric model, and an Exponentially Weighted Moving Average (EWMA) statistic is derived therefrom. The EWMA statistic is able to identify whether the machine is in a normal state ("in control") or in an abnormal state ("out of control"). Additionally, an EWMA control chart is generated that distinguishes between normal and abnormal conditions, and between different abnormal conditions. As a result, once the EWMA statistic is generated, it is compared to the EWMA chart for determination of the specific fault that is ailing the machine.

Claims

1. A computer-based method for monitoring and diagnosing an abnormal machine condition, comprising the steps of:

(1) operating a machine under normal machine conditions;

(2) defining a parametric model for said machine operating under said normal machine condition;

(3) calculating an Exponentially Weighted Moving Average (EWMA) statistic based on prediction errors generated in fitting abnormal vibration signals to said parametric model; and

(4) comparing said EWMA statistic to a limit, wherein if said EWMA statistic is above said limit said machine is operating abnormally.

2. The computer-based method of claim 1, further comprising the step of comparing said EWMA statistic to different values in a chart to determine a specific abnormal condition affecting said machine.

3. A computer-based method for monitoring and diagnosing a machine condition, comprising the steps of:

(1) operating a machine under normal machine conditions;

(2) collecting a first set of data from said machine, wherein said first set of data is indicative of the operation of said machine under said normal machine conditions;

(3) selecting an AR order for said normal machine condition, and generating an AR model for said first set of data, wherein said AR model has a first order parameters (φ

_i1), second order parameters (φ

_i2), up to p order parameters (φ

_ip) for i=1, 2, . . . , n data sets;

(4) calculating an average value for said first order AR parameter through said pth order AR parameter from said AR models in order to define a normal model that is representative of said normal machine condition;

(5) collecting a second set of data from a machine under diagnosis, wherein said second set of data is representative of an abnormal machine condition;

(6) fitting said second set of data representative of said abnormal machine condition to said normal model to generate a fitted model, wherein said fitted model is an indicator of how closely said normal model fits said second set of data;

(7) calculating forward and backward prediction errors to determine a ρ

_normalized^fb value for said second set of data;

(8) calculating an exponentially weighted moving average (EWMA) statistic based on said ρ

_normalized^fb value, wherein said EWMA statistic is an indicator of the overall machine condition;

(9) comparing said EWMA statistic to an upper control limit to determine if said machine under diagnosis is in a state of control or is a state of out-of-control, wherein if said EWMA statistic exceeds said upper control limit this is a signal that an abnormal machine condition exists in said machine under diagnosis.

4. The computer-based method of claim 3, further comprising the step of determining an appropriate AR model using a modified covariance method, final prediction error, Akaike information criteria, and criterion autoregressive transfer function, wherein said AR model adequately describes a normal machine condition.

5. The computer-based method of claim 3, wherein said EWMA is calculated with λ

=0.7.