Method to extract parameters from in-situ monitored signals for prognostices

• US 20100100337A1
• Filed: 10/13/2009
• Published: 04/22/2010
• Est. Priority Date: 10/16/2008
• Status: Active Grant

• Current Assignee: Oxfordian
• Original Assignee: Oxfordian
• Inventors: Pecht, Michael G., Vichare, Nikhil M.
• Granted Patent: US 8,521,443 B2
• Time in Patent Office: Days
• US Class Current: 702/34
• CPC Class Codes: G06F 11/008 Reliability or availability analysis

Abstract

Through the methods of this invention, prognostics tools are provided to more efficiently and more accurately predict when a component product may fail. In a first embodiment a method is described whereby a number of environmental factors are monitored, the provided sensors obtaining raw data, variations in said raw data measured, the parameters of interest extracted and binned according to predetermined criteria, with the raw data thereafter discarded. In a second embodiment of the invention, the same sensor readings are recorded along with the performance characteristics for the component/product and compared to estimated performance for the product. The performance drift is then observed, whereby in monitoring said drift, trends may be determined and time or cycles to failure predicted.

Claims

1. A method for analyzing the environmental and operational exposures of a product or article of manufacture, whereby:

(a) irregular changes over time in the environmental and operational condition are first detected and recorded as raw data;

(b) cyclic range, cyclic mean, ramp rates, dwell times, dwell loads and their correlations are then extracted from said recorded raw data and stored;

(c) the individual pieces of recorded raw data are thereafter discarded;

whereby using such parameter extraction process, data storage reductions of up to 99.03% achieved for a given processing duration.

2. The method of claim 1 in which:

(a) optimal bin-widths are calculated using historic sensor data to reduce the error between the true density and the estimated density of the load values; and

,

(b) the load parameters are further stored in bins of optimal bin width to achieve further reductions in data storage.

3. The method of claim 1 in which the load parameters are used by a failure model to assess the health of the product or article of manufacture and predict its remaining life.

4. A computer assisted method for collecting and storing load parameter data such as environmental and operational exposure data, and thereafter analyzing such collected data whereby said data is discarded following analysis and storage of the results, the method comprising:

Selecting an article of manufacture to be monitored;

associating one or more sensors with said article of manufacture, whereby said one or more sensors are capable of sensing environmental or operational load conditions to which the article of manufacture is subjected to over time;

collecting raw sensor data over time and storing the said collected load data in an electronic memory, the interval of time between the taking of sensor readings predetermined;

periodically subjecting the stored data to a data reduction process, in which the data is converted to other values, and the raw data then discarded, said data reduction process including the steps of;

processing, with the aid of a computer, the raw data which has been stored in memory, using a moving average filter to remove noise;

conducting a computer directed scan of the processed data to identify those collection of data points where the sensed load condition is either continuously increasing, decreasing or dwells at a constant value;

identifying all dwell time data regions, and storing same in the computer memory as a matrix of dwell times along with the associated constant load level data for each dwell region;

processing the remaining raw data sets which now consists of continuously increasing and decreasing data sets, with the aid of a computer, to identify full and half cycles, identifying the cyclic range for each identified cycle, and calculating the mean load, ramp-up rate and ramp-down rate for each cyclic load;

storing in electronic memory the identified cyclic range, and calculated mean load, and ramp up and down rates for each cycle;

performing a correlation calculation whereby dwell time vs. load at dwell, mean load and ramp rates are calculated and stored in electronic memory;

wherein, previously stored raw data is thereafter deleted.

5. The computer assisted method of claim 4 wherein said identified and calculated stored load parameter values are thereafter stored in data bins based on optimal bin widths, both the bin widths and the frequency of stored load parameter values assigned to each bin, recorded and stored in said electronic memory.

6. The computer assisted method of claim 5 wherein said optimal bin width is based on estimates of standard deviation and sample size from historic sensor data.

7. The computer assisted method of claim 4 wherein the one or more sensors are embedded in the article of manufacture.

8. The computer assisted method of claim 4 wherein the one or more sensors are autonomous sensors retrofitted to the article of manufacture.

9. The computer assisted method of claim 4 wherein a predetermined failure model exists for the article of manufacture.

10. The computer assisted method of claim 4 wherein the predetermined interval of time for recording a raw data sensor reading is selected from the group comprising seconds, fractions of a second, minutes, hours, and days.

11. The computer assisted method of claim 4 wherein the at least one of said load parameters is temperature.

12. The computer assisted method of claim 4 wherein, in addition to the identified dwell regions of constant load level, data regions of small cycles with a constant or near constant mean and data regions of relatively large cycle with an extremely slow ramp rate are classified as dwell regions, and stored as dwell times along with the associated mean load value of such classified regions this data stored along with the matrix of dwell times for constant load dwells.

13. The computer assisted method of claim 4 wherein the correlation calculation of dwell time vs. load at dwell is performed at the time that the dwell time data is identified and stored, and the raw dwell time data is then discarded prior to performing the next data reduction step.

14. An electronic module for monitoring the environmental conditions to which an article of manufacture is exposed, and predicting when that component might fail, said module comprising:

(a) one or more environmental, operational or performance sensors;

(b) means for periodically recording the sensor readings;

(c) means for processing said sensor readings whereby based on the sensor readings, load parameters are calculated;

(d) means for binning the calculated load parameters into bins of optimal bin-widths; and

,

(e) means for accumulating the damage conditions to predict remaining useful life based on current conditions and predictions of future use based on historic load data.

15. The electronic module of claim 14 wherein the environmental sensors include one or more sensors to monitor conditions selected from the group comprising strain, temperature, acceleration, and humidity.

16. A method for predicting when a product might fail during long term usage, said method including the steps of:

(a) continuously monitoring in-situ of various environmental and/or operational parameters;

(b) monitoring one or more product performance parameters;

(c) Modeling the relationship between the environmental and/or operation parameters and the performance parameters;

(d) determining the performance drift of the product by calculating the difference between the performance parameter estimated by the model and the actual performance parameter observed;

(e) analyzing statistical features from the distributions of performance drift;

(f) analyzing the trend of said statistical features to determine the onset of degradation and assessing the health of the product; and

,

(g) thereafter trending the said statistical features to project when the trend line will meet the defined failure criteria, and thus estimate remaining product life.

17. A computer assisted method of failure prediction of an article of manufacture in use and exposed to various load parameters including environmental and operational parameters, including the steps of:

selecting an article of manufacture to be monitored for health assessment;

selecting one or more load factors to be monitored as well as a performance parameter to be monitored;

monitoring and recording the selected performance parameter and the one or more load factors on a periodic basis, storing the monitored data in an electronic memory;

using a predetermined operational model for the selected article of manufacture, electronically, calculating an expected value of the performance parameter at each point in time that the preselected load factor is sampled, and storing said calculated value in electronic memory;

electronically comparing the monitored value of the selected performance parameter to the calculated expected value, the difference between the calculated and expected values representing the degree of performance parameter drift, which drift value is stored in electronic memory;

periodically electronically subjecting the drift values to one or more statistical feature analysis calculations,

using the results of the feature analysis, electronically extrapolating the drift trend over time to that time at which the article of manufacture, by definition, is deemed to have reached the point of failure; and

,

thereafter determining from said extrapolated data if the article of manufacture is likely to fail earlier or later than the predicative model for said article.

18. The method of failure prediction of claim 17 wherein at said predefined time of failure, a failure threshold for the performance parameter being monitored is also defined.

19. The method of failure prediction of claim 18 wherein failure of the article of manufacture is determined to be that time at which the trend line of the performance parameter indicates that the failure threshold for that parameter will be reached.

20. The method of failure prediction of claim 17 wherein a warning is issued if the extrapolated data indicates a failure of the article prior to the time of defined failure.

21. The method of failure prediction of claim 17 wherein the selected statistical analysis method applied is the 95% cumulative peaks analysis.

22. The method of failure prediction of claim 17 wherein the load values and performance parameters, as raw data, are sampled and periodically stored.

23. The method of failure prediction of claim 22 wherein after a set number of sampling periods have occurred, a feature analysis is performed on said stored data, an extrapolation of the results of the analysis performed, and the stored raw data used for the said calculations then discarded.