Machine fault diagnostics system and method

• US 5,566,092 A
• Filed: 12/30/1993
• Issued: 10/15/1996
• Est. Priority Date: 12/30/1993
• Status: Expired due to Term

• Current Assignee: Caterpillar Incorporated
• Original Assignee: Caterpillar Incorporated
• Inventors: Lin, Shui-Shun, Huang, Hsin-Hao, Knapp, Gerald M., Wang, Hsu-Pin, Lin, Chang-Ching, Spoerre, Julie K.
• Primary Examiner(s): Voeltz, Emanuel T.
• Assistant Examiner(s): WACHSMAN, HAL D
• Application Number:
  US08/176,482
• Time in Patent Office: 1,020 Days
• Field of Search: 364/131, 364/164, 364/474.01, 364/474.11, 364/474.15-474.17, 364/505, 364/506, 364/550, 364/551.01, 364/551.02, 364/579, 364/578, 395/3, 395/11, 395/21, 395/22, 395/50, 395/66, 395/75, 395/77, 395/82-84, 395/88, 395/93, 395/97, 395/900, 395/903, 395/904, 395/906, 395/907, 395/909, 395/911, 395/912, 395/914 , 395/915, 395/932, 395/2.25-2.28, 395/2.11
• US Class Current: 702/185
• CPC Class Codes: G01N 2291/0258 Structural degradation, e.g. fatigue of composites, ageing of oils
• CPC Class Codes: G01N 29/043 in the interior, e.g. by shear waves
• CPC Class Codes: G01N 29/14 using acoustic emission techniques echo of particles G01N29/046; measuring mechanical vibrations or acoustic waves in solids in general G01H1/00
• CPC Class Codes: G01N 29/4472 Mathematical theories or simulation
• CPC Class Codes: G01N 29/4481 Neural networks
• CPC Class Codes: G01N 29/449 Statistical methods not provided for in G01N29/4409, e.g. averaging, smoothing and interpolation
• CPC Class Codes: G05B 19/4184 characterised by fault tolerance, reliability of production system
• CPC Class Codes: G05B 2219/33027 Artificial neural network controller
• CPC Class Codes: G05B 2219/33303 Expert system for diagnostic, monitoring use of tree and probability
• CPC Class Codes: G05B 2219/34065 Fuzzy logic, controller
• CPC Class Codes: G06F 18/23211 with adaptive number of clusters
• CPC Class Codes: G07C 3/00 Registering or indicating the condition or the working of machines or other apparatus, other than vehicles
• CPC Class Codes: Y02P 90/02 Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
• CPC Class Codes: Y10S 706/904 Manufacturing or machine, e.g. agricultural machinery, machine tool
• CPC Class Codes: Y10S 706/912 Manufacturing or machine, e.g. agricultural machinery, machine tool

Abstract

The invention provides a machine fault diagnostic system to help ensure effective equipment maintenance. The major technique used for fault diagnostics is a fault diagnostic network (FDN) which is based on a modified ARTMAP neural network architecture. A hypothesis and test procedure based on fuzzy logic and physical bearing models is disclosed to operate with the FDN for detecting faults that cannot be recognized by the FDN and for analyzing complex machine conditions. The procedure described herein is able to provide accurate fault diagnosis for both one and multiple-fault conditions. Furthermore, a transputer-based parallel processing technique is used in which the FDN is implemented on a network of four T800-25 transputers.

Claims

1. A fault diagnostic system, comprising:

(a) a data acquisition module that collects sensory signals;

(b) a diagnostic module, connected to said data acquisition module, that performs on-line fault detection for a physical machine or process, fault diagnostics, and provides recommendations in regard to said on-line fault detection and said fault diagnostics; and

(c) a machine model module, connected to said diagnostic module, that provides a physical model for identifying fault conditions that cannot be diagnosed by said diagnostic module.

2. The fault diagnostic system of claim 1 , wherein said data acquisition module includes a plurality of sensors each capable of real-time data acquisition.

3. The fault diagnostic system of claim 1, wherein said diagnostic module includes a parametric modeling module, a modified ARTMAP neural network, a fuzzy logic module, and an expert system.

4. The fault diagnostic system of claim 3, wherein said machine model module includes physical models for bearing and gears, and provides data for preliminary training of said modified ARTMAP neural network.

5. The fault diagnostics system of claim 1, wherein said data acquisition module includes means for preprocessing said sensory signals using an autoregressive model.

6. The fault diagnostic system of claim 1, wherein said diagnostic module comprises detection means for performing fault detection that includes detecting abnormal conditions in said physical machine or process by using an overall root mean square (RMS) measurement and a covariance statistic of an exponentially weighted moving average method (EWMA), wherein a control limit is set for said RMS and said EWMA, and if said sensory conditions exceeds said control limit than an abnormal condition exists.

7. The fault diagnostic system of claim 1, wherein said diagnostic module comprises identification means for performing fault identification based on a model-based reasoning approach that includes fuzzy logic methodology.

8. The fault diagnostic system of claim 1, wherein said diagnostic module comprises an expert system that performs fault verification and that provides recommendations about detected faults to a user.

9. The fault diagnostic system of claim 1, wherein said diagnostic module and said machine model module are implemented on a transputer network.

10. The fault diagnostic system of claim 1, wherein said sensory signals include vibration signals, pressure signals, and/or temperature signals from said physical machine or process.

11. The fault diagnostic system of claim 1, wherein said diagnosis module comprises a fault diagnostic network, wherein said fault diagnostic network includes:

(a) an ART module that accepts an input pattern, said ART module having a first layer and a second layer, and configured to perform a first vigilance test and generate a recognition category; and

(b) a map field, connected to said ART module and to a target output pattern, that performs a mapping between said recognition category and said target output pattern, and triggers the performance of a second vigilance test;

wherein said second vigilance test determines the closeness between said target output pattern and said recognition category.

12. The fault diagnostic system of claim 11, wherein said first layer comprises three levels that each include:

means for performing integration of intrafield and interfield inputs to produce an integrated activation signal; and

means for performing normalization of said integrated activation signal.

13. The fault diagnostic system of claim 11, wherein said ART module is an ART2 network that can accept binary or analog input patterns.

14. The fault diagnostic system of claim 11, wherein said target output pattern is a binary vector, wherein each node of said binary vector corresponds to a particular machine or process condition.

15. The fault diagnostic system of claim 11, further comprising parameter means for fitting a raw vibration signal collected from a machine under review to a predetermined parametric model, wherein said parameter means generates a parameter.

16. The fault diagnostic system of claim 15, further comprising a normalization module, connected to said parameter means, that removes meaningful negative values that said ART module is not able to recognize from said parameter.

17. The fault diagnostic system of claim 16, wherein said normalization module comprises:

(1) means for dividing said parameter into a negative part and a positive part;

(2) means for scaling said negative part and said positive part by dividing said parameter by a maximum parameter value.

18. The fault diagnostic system of claim 11, wherein each node of said second layer corresponds to a particular fault condition.

19. The fault diagnostic system of claim 1, wherein said machine model module further provides a deep fault reasoning mechanism to identify complex or multiple fault conditions.

20. A method for diagnosing a physical machine or process, the method comprising the steps of:

(1) acquiring a first set of data from the physical machine or process;

(2) preprocessing said first set of data using an autoregressive model, said preprocessing generating an autoregressive parameter; and

(3) detecting abnormal conditions in said autoregressive parameter using an overall root means square (RMS) measurement and a covariance statistic of an exponentially weighted moving average (EWMA), wherein if an abnormal condition is detected then,

(a) identifying whether said physical machine or process has a fault, including,

(i) determining a hypothesis with the aid of a fault diagnostic network, and if said fault diagnostic network cannot generate a hypothesis, then

(ii) determining a hypothesis with the aid of a model-based reasoning approach, wherein said model-based reasoning approach uses fuzzy logic; and

(b) supplying said identifiable fault to a fault expert system having a knowledge base with a set of rules, wherein said fault reasoning expert system checks said identifiable fault against said set of rules.

21. The method of claim 20, wherein step (3) comprises the steps providing an indication to a user that said machine or process is operating in a normal manner, a potentially problematic manner, or a dangerous manner.

22. The method of claim 20, wherein said fault reasoning expert system also checks said identifiable fault against damage and repair history and machine usage information.

23. The method of claim 22, further comprising the step of displaying said identifiable fault along with recommendations via a user interface.

24. The method of claim 20, further comprising the steps of preprocessing said second set of acquired data using an autoregressive model, said preprocessing generating an autoregressive parameter, and normalizing said autoregressive parameter.

25. The method of claim 20, further comprising the step of acquiring a second set of data from said physical machine or process after said abnormal condition is detected.