Condition-based prognosis for machinery

• US 6,411,908 B1
• Filed: 08/02/2000
• Issued: 06/25/2002
• Est. Priority Date: 04/27/2000
• Status: Expired due to Fees

• Current Assignee: Progressive Machinery Corporation
• Original Assignee: Progressive Machinery Corporation
• Inventors: Talbott, Carlos M.
• Primary Examiner(s): Hilten, John S.
• Assistant Examiner(s): VO, HIEN XUAN
• Application Number:
  US09/630,985
• Time in Patent Office: 692 Days
• Field of Search: 702/34, 702/33, 702/35, 702/81, 702/82, 702/181, 702/182, 702/183, 702/184, 702/185, 73/530.2, 731/181, 731/173, 737/87, 706/45, 706/912, 706/56, 700/79, 700/292, 700/21
• US Class Current: 702/34
• CPC Class Codes: G05B 23/0224 Process history based detection method, e.g. whereby history implies the availability of large amounts of data
• CPC Class Codes: G07C 3/00 Registering or indicating the condition or the working of machines or other apparatus, other than vehicles

Abstract

A method for assembling condition monitoring histories of same-type machines that have lived in same-type environments and have failed as a result of the same failure mode, estimating the remaining life with confidence bounds in an operating machine that presents a set of condition symptoms over time and that is diagnosed with a pending failure mode, and deciding when to replace/repair an operating machine (diagnosed with a specific failure mode condition) based on the cost of its estimated performance over its predicted remaining life.

Claims

1. A method for estimating the remaining life in an operating machine comprising:

accumulating symptom data, condition data and time stamp data from machines the same or similar to the operating machine, the symptom data and the condition data measured and recorded at various times t_i, including time of failure t₀ and failure mode;

calculating remaining life values associated with each set of accumulated symptom data and condition data recorded for each machine of the machines;

assembling the symptom data, the condition data, and the remaining life values into prognostic knowledge-bases according to machine type, operating environment, and failure mode;

storing the prognostic knowledge-bases in matrix arrays to facilitate mathematical analyses for the purpose of predicting remaining life in the operating machine running in a certain environment, diagnosed with a certain failure mode, and presenting a set of symptom data (s*) and condition data; and

analyzing data in a prognostic knowledge-base of the prognostic knowledge-bases appropriate to the operating machine for the purpose of estimating the remaining life in the operating machine and determining a statistical confidence bounds around the estimate of remaining life.

2. The method of claim 1 wherein accumulating symptom data and condition data further comprises the steps of:

a. identifying sets of symptom variables germane to particular failure modes for the machines same or similar to the operating machine;

b. defining consensus states of machine failure;

c. defining consensus failure modes for the machines such that a primary cause of failure can be determined accurately and consistently; and

d. acquiring the symptom data along with time-stamp identification as to when each set of symptom data was acquired, when the machines had failed, and what was the primary cause of failure.

3. The method of claim 1 wherein calculating remaining life values associated with each set of accumulated symptom data and condition data recorded for a given machine of the machines further comprises the steps of:

a. subtracting an operating time-stamp value made when a set of symptom data and condition data were measured from a failure time-stamp value made at the time of machine failure resulting in a remaining life value associated with each set of symptom data and condition data; and

b. translating the remaining life value into units of measure that are common to those of other same-type machines.

4. The method of claim 1 wherein assembling the symptom data, the condition data, and the remaining life values into prognostic knowledge-bases further comprises the steps of:

a. identifying a number m of different symptom variables and a number z of different condition variables within a single machine history that are appropriate to its primary failure mode as determined during that machine'"'"'s autopsy; and

b. creating a machine history matrix with (m+z+1) columns and n rows and populating the machine history matrix with the symptom data, the condition data and the remaining life values.

5. The method of claim 4 further comprising the step of:

c. combining a number p of all machine history matrices, for machines the same or similar to the operating machine and failed according to the same or similar failure mode, into the prognostic knowledge-base matrix array wherein symptom data, condition data, and remaining life values are aligned by column within the knowledge-base matrix array with any empty column cells filled with zeroes.

6. The method of claim 1 wherein storing the prognostic knowledge-bases in matrix arrays for the purpose of predicting remaining life in the operating machine further comprises the steps of:

a. transferring the prognostic knowledge-bases matrix arrays into electronic media suitable for rapid retrieval and analysis via computer programs; and

b. updating the prognostic knowledge-bases matrix arrays by appending a newly acquired machine history matrix to its appropriate prognostic knowledge-base matrix array as another layer in the array.

7. The method of claim 1 wherein analyzing data in the prognostic knowledge-base further comprises the steps of:

a. extracting those cell values of symptom variables and their corresponding remaining life found on a row, k, within a layer of the prognostic knowledge-base matrix array such that the symptom data found on that row are in a “

smallest neighborhood”

around s*, such extraction of single doublets (s_i^T, r_i) occurring for every layer in the matrix array;

b. determining those layers of the prognostic knowledge-base matrix array that correspond to a machine that failed prior to attaining any one of the symptom value levels described in s*, the number of said layers is q out of p total layers, and the remaining life value in each extracted row of these q layers is set at zero;

c. computing the conditional probability of observing in the matrix array a non-zero remaining life value extracted per Step a. above as {(1/p)/((p−

q)/p)};

d. computing the conditional expected value of remaining life by multiplying each non-zero extracted remaining life value per Step a. by its conditional probability calculated per Step c. and summing each multiplication;

e. computing the conditional variance of remaining life by multiplying the square of each non-zero extracted remaining life value per Step a. less the conditional expected value of remaining life per Step d. by its conditional probability calculated per Step c. and summing each multiplication; and

f. computing an estimate of remaining life with confidence bounds as the conditional expected value of remaining life per Step d. plus and minus an appropriate Standard Normal density value, Z, multiplied by the square root of the conditional variance per Step e. and divided by the square root of the number p of total layers in the matrix array.

8. A method for determining when to replace an operating machine comprising:

accumulating symptom data, condition data and time stamp data from machines the same or similar to the operating machine, the symptom data and the condition data measured and recorded at various times t_i, including time of failure t₀ and failure mode;

calculating remaining life values associated with each set of accumulated symptom data and condition data recorded for each machine of the machines;

assembling the symptom data, the condition data, and the remaining life values into prognostic knowledge-bases according to machine type, operating environment and failure mode;

storing the prognostic knowledge-bases in matrix arrays to facilitate mathematical analyses for the purpose of estimating the future performance output in the operating machine running in a certain environment, diagnosed with a certain failure mode, and presenting a set of symptom data (s*) and condition data; and

analyzing data in a prognostic knowledge-base of the prognostic knowledge-bases appropriate to the operating machine for the purpose of estimating a performance output over an expected lifetime of the operating machine.

9. The method of claim 8 wherein accumulating symptom data and condition data further comprises the steps of:

a. identifying sets of symptom variables germane to particular failure modes for the machines same or similar to the operating machine;

b. identifying sets of condition variables for the machines that may have some effect on the deterioration time frame for a particular failure mode as evidenced by certain symptom variables or that are affected themselves by the degree of deterioration for a particular failure mode;

c. defining consensus states of machine failure;

d. defining consensus failure modes for the machines such that a primary cause of failure can be determined accurately and consistently; and

e. acquiring the symptom data and the condition data along with time-stamp identification as to when each set of symptom data and condition data was acquired, when the machines had failed, and what was the primary cause of failure.

10. The method of claim 8 wherein calculating remaining life values associated with each set of accumulated symptom data and condition data recorded for a given machine of the machines further comprises the steps of:

a. subtracting an operating time-stamp value made when a set of symptom data and condition data were measured from a failure time-stamp value made at the time of machine failure resulting in a remaining life value associated with each set of symptom data and condition data; and

b. translating the remaining life value into units of measure that are common to those of other same-type machines.

11. The method of claim 8 wherein assembling the symptom data, the condition data, and the remaining life values into prognostic knowledge-bases further comprises the steps of:

a. identifying a number m of different symptom variables and a number z of different condition variables within a single machine history that are appropriate to its primary failure mode as determined during that machine'"'"'s autopsy; and

b. creating a machine history matrix with (m+z+1) columns and n rows and populating the machine history matrix with the symptom data, the condition data and the remaining life values.

12. The method of claim 11 further comprising the step of:

c. combining a number p of all machine history matrices, for machines the same or similar to the operating machine and failed according to the same or similar failure mode, into the prognostic knowledge-base matrix array wherein symptom data, condition data, and remaining life values are aligned by column within the knowledge-base matrix array with any empty column cells filled with zeroes.

13. The method of claim 8 wherein storing the prognostic knowledge-bases in matrix arrays for the purpose of predicting the performance output over the lifetime of the operating machine further comprises the steps of:

a. transferring the prognostic knowledge-bases matrix arrays into electronic media suitable for rapid retrieval and analysis via computer programs; and

b. updating the prognostic knowledge-bases matrix arrays by appending a newly acquired machine history matrix to its appropriate prognostic knowledge-base matrix array as another layer in the array.

14. The method of claim 8 wherein estimating the output over an expected lifetime of the operating machine further comprises the steps of:

a. identifying subsets, e_i, of condition variables that are associated with a certain machine'"'"'s performance and determining the relationship of a performance output d_i in terms of performance related variables;

b. extracting those cell values of symptom variables, their corresponding performance related output values, and their corresponding remaining life values found on and below a certain row, k, within a layer of an appropriate matrix array such that the symptom values found on that row are in the “

smallest neighborhood”

around s*, such extraction of multiple triplets (s_i^T, d_i^T, r_i) occurring for each layer in the matrix array;

c. determining those layers of the matrix array that correspond to a machine that failed prior to attaining any one of the symptom value levels described in s*, the number of the layers is q out of p total layers, and the remaining life value in each extracted triplet of these q layers is set at zero;

d. calculating, for each layer, the sum of multiplications of the difference d_k+i minus d<sub>k+i−

1</sub> times the value r_k+i for i=0, 1, 2, . . . , ∞

;

e. computing the conditional probability of observing in any layer of the matrix array a non-zero sum calculated per Step d. above as {(1/p)/((p−

q)/p)}; and

f. computing the conditional expected value of performance output over the conditional expected remaining life by multiplying each non-zero sum per Step d. by its conditional probability calculated per Step e. and summing each multiplication.

15. The method of claim 8 wherein deciding when to replace the operating machine based upon the estimate of remaining life further comprises the steps of:

a. determining the expected performance output of a healthy machine over the expected lifetime of the operating machine and converting said performance value into monetary units;

b. determining the economic value of expected remaining life in the operating machine;

c. computing the sum of the monetary values found in Steps a and b; and

d. comparing at least one of a future economic impact and a future performance impact of the operating machine with the result of Step c to determine whether to replace the operating machine.

16. A method for estimating the remaining life in an operating machine comprising:

accumulating symptom data, condition data and time stamp data at which the symptom data and the condition data was acquired, time of failure data, and failure mode data, from machines the same or similar to the operating machine;

calculating remaining life values associated with each set of accumulated symptom data and condition data from the time stamp data and time of failure recorded for the machines;

assembling the symptom data, the condition data, and the remaining life values into a prognostic knowledge-base matrix array; and

analyzing the prognostic knowledge-base matrix array to estimate the remaining life in the operating machine with a statistical confidence bounds.

17. The method of claim 16 wherein calculating remaining life values associated with each set of accumulated symptom data, condition data and time data recorded for a given machine of the machines further comprises the step of:

determining each remaining life value of the remaining life values from an operating time-stamp value made when a set of symptom data and condition data were measured and a failure time-stamp value made at the time of machine failure.

18. The method of claim 16 wherein assembling the symptom data, the condition data, and the remaining life values into prognostic knowledge-bases further comprises the steps of:

creating a machine history matrix with (m+z+1) columns, where there are m different symptom variables and z different condition variables within a single machine history, and n rows and populating the machine history matrix with the symptom data, the history data and the remaining life values; and

combining a number p of all machine history matrices into the prognostic knowledge-base matrix array wherein symptom data and condition data are aligned by column within the knowledge-base matrix array.

19. The method of claim 16 wherein analyzing data in the prognostic knowledge-base further comprises the steps of:

a. extracting values of symptom variables and their corresponding remaining life found in the prognostic knowledge-base matrix array;

b. computing a conditional probability of observing, in the matrix array, remaining life values extracted per Step a. above;

c. computing a conditional expected value of remaining life;

d. computing a conditional variance of remaining life; and

f. computing an estimate of remaining life with statistical confidence bounds.

20. A method for determining when to replace an operating machine comprising:

accumulating symptom data, condition data and time stamp data at which the symptom data and the condition data was acquired, from machines the same or similar to the operating machine;

calculating remaining life values associated with each set of accumulated symptom data, condition data and time stamp data recorded for the machines;

assembling the symptom data, the condition data, and the remaining life values into a prognostic knowledge-base matrix array; and

analyzing the prognostic knowledge-base matrix array to estimate the performance output over the expected lifetime of the operating machine.

21. The method of claim 20 wherein calculating remaining life values associated with each set of accumulated symptom data and condition data recorded for a given machine of the machines further comprises the step of:

determining each remaining life value of the remaining life values from an operating time-stamp value made when a set of symptom data and condition data were measured and a failure time-stamp value made at the time of machine failure.

22. The method of claim 20 wherein assembling the symptom data, the condition data, and the remaining life values into prognostic knowledge-bases further comprises the steps of:

creating a machine history matrix with (m+z+1) columns, where there are in different symptom variables and z different condition variables within a single machine history, and n rows and populating the machine history matrix with the symptom data, the condition data and the remaining life values; and

combining a number p of all machine history matrices into the prognostic knowledge-base matrix array wherein symptom data, condition data, and remaining life values are aligned by column within the knowledge-base matrix array.

23. The method of claim 20 wherein storing the prognostic knowledge-bases in matrix arrays for the purpose of predicting performance output over the lifetime of the operating machine further comprises the steps of:

a. transferring the prognostic knowledge-bases matrix arrays into electronic media suitable for rapid retrieval and analysis via computer programs; and

b. updating the prognostic knowledge-bases matrix arrays by appending a newly acquired machine history matrix to its appropriate prognostic knowledge-base matrix array as another layer in the array.

24. The method of claim 20 wherein estimating at least one of the future economic impact and the future performance impact of the operating machine further comprises the steps of:

a. identifying subsets of condition variables that are associated with a certain machine'"'"'s performance and determining the relationship of a performance output in terms of performance related variables;

b. extracting those cell values of symptom variables, their corresponding performance related variables, and their corresponding remaining life value occurring for each layer in the prognostic knowledge-base matrix array; and

c. computing the conditional expected value of performance output over the conditional expected remaining life.

25. The method of claim 20 wherein deciding when to replace the operating machine based upon the estimate of its remaining life further comprises the steps of:

a. determining the expected performance output of a healthy machine over the expected lifetime of the operating machine and converting said performance value into monetary units;

b. determining the economic value of expected remaining life in the operating machine;

c. computing the sum of the monetary values found in Steps a and b; and

d. comparing at least one of the future economic impact and the future performance impact of the operating machine with the result of Step c to determine whether to replace the operating machine.

26. A method for estimating the remaining life in a generic operating machine diagnosed with a generic failure mode and presenting symptoms of a given surface temperature and a given vibration velocity comprising the steps of:

accumulating symptom data, condition data, and time stamp data from machines the same or similar to the operating machine, the symptom data and the condition data measured and recorded at various times t_i and the time of failure t₀ and failure mode also recorded as in a Machine History Table;

calculating remaining life values associated with each set of accumulated symptom data and condition data recorded for each of the failed machines from the available time stamp data;

assembling the symptom data, the condition data, and the remaining life values into a prognostic knowledge-base matrix array; and

analyzing the prognostic knowledge-base matrix array to estimate the remaining life in the generic operating machine with a predetermined statistical confidence bounds.

27. A method for determining when to replace a generic operating machine diagnosed with a generic failure mode and presenting symptoms of a given surface temperature and a given vibration velocity comprising:

accumulating symptom data, condition data, and time stamp data from machines the same or similar to the operating machine, the symptom data and the condition data measured and recorded at various times t_i and the time of failure t₀ and failure mode also recorded;

calculating remaining life values associated with each set of accumulated symptom data and condition data recorded for each of the failed machines from the available time stamp data;

assembling the symptom data, the condition data, and the remaining life values into a prognostic knowledge-base matrix array;

analyzing the prognostic knowledge-base matrix array to estimate the performance output over the expected remaining life of the operating machine; and

comparing the estimated performance output over the expected remaining life of the operating machine against the expected performance output of a healthy machine over the same time horizon plus the economic value of expected remaining life in the operating machine.