Component failure prediction

Patent metadata

Field	Value
Publication number	US-11003518-B2
Application number	US-201616334857-A
Country	US
Kind code	B2
Filing date	Sep 29, 2016
Priority date	Sep 29, 2016
Publication date	May 11, 2021
Grant date	May 11, 2021

Abstract

Example systems may relate to component failure prediction. A non-transitory computer readable medium may contain instructions to analyze a plurality of features corresponding to a component of a system. The non-transitory computer readable medium may further contain instructions to determine which of the plurality of features to use to model a failure of the component. The non-transitory computer readable medium may contain instructions to generate a plurality of models to model the failure of the component and assemble the plurality of models into a single model for predicting component failure. The non-transitory computer readable medium may further contain instructions to extract data associated with a component failure predicted by the single model and correlate the data associated with the predicted component failure with the single model.

First claim

What is claimed:

1. A non-transitory computer readable medium containing instructions that when executed by a processor cause the processor to:
collect data from a sensor monitoring a component of a system, wherein the component is one of an electrical component and a mechanical component;
analyze a plurality of features, wherein each of the features corresponds to the data collected from the sensor;
determine which of the plurality of features to use to generate failure models of the component based on a correlation to a first failure type, wherein features with a first correlation above a threshold correlation are utilized to generate the failure models and features with a second correlation below the threshold correlation are discarded;
generate a plurality of models to model the first failure type of the component;
assemble the plurality of models into a single model for predicting a failure of the first failure type of the component;
extract data associated with the failure predicted by the single model;
correlate data associated with a detected failure with the single model; and
determine when the plurality of features has successfully predicted the failure of the first failure type of the component.

2. The non-transitory computer readable medium of claim 1 , wherein the instructions to assemble the plurality of models include instructions to assemble the plurality of models based on a determined range of failures for the component.

3. The non-transitory computer readable medium of claim 1 , wherein the instructions to determine which of the plurality of features to be used include instructions to select a portion of the plurality of features based on a determined quantity of features for generating the plurality of models.

4. The non-transitory computer readable medium of claim 1 , further comprising instructions to:
predict, based on the single model, the failure of the component; and
transmit an alert that the component is predicted to fail.

5. The non-transitory computer readable medium of claim 1 , comprising instructions to:
determine whether each of the plurality of features has successfully predicted a failure of the component;
calculate a correlation between each feature and its successful prediction of component failure; and
determine a rate at which each of the plurality of features fails to successfully predict component failure.

6. A system, comprising:
a data collector to collect data from a plurality of sensors monitoring a component of a system, wherein the component is one of an electrical component and a mechanical component;
a feature analyzer to analyze a plurality of features based on the data collected from the plurality of sensors;
a feature composer to determine and select which of the plurality of features to use based on a correlation to a failure type, wherein features with a first correlation above a threshold correlation are selected to generate failure models of the component and features with a second correlation below the threshold correlation are discarded;
a model generator to generate a plurality of failure models to model the first failure type of the component based on the selected features;
a model composer to assemble the plurality of failure models into a single failure model for predicting a failure of the failure type of the component by:
receiving an output from the plurality of generated failure models; and
combining the output from the plurality of generated failure models into a single output;

an extractor to extract data associated with the failure predicted by the single model; and
a correlator to correlate data associated with a detected failure with the single failure model; and determine when the plurality of features has successfully predicted the failure of the failure type of the component.

7. The system of claim 6 , further comprising the failure extractor to:
determine that a component failure has occurred;
extract data corresponding to the failed component;

generate association information for the component based on the corresponding data; and
provide the association information to the feature generator.

8. The system of claim 6 , further comprising the model composer to:
determine an aspect of data represented by each of the plurality of failure models;
analyze a combination of the plurality of failure models to determine a coverage of the aspect by the combination of the failure models;
determine that a combination of the plurality of failure models provides a coverage of the aspect above a threshold; and
select the combination of the plurality of failure models.

9. The system of claim 6 , further comprising the feature composer to:
determine, based on the analysis of the plurality of features, that a feature has a correlation to a failure that is below a threshold correlation; and
discard the feature based on the determination that the correlation is below the threshold correlation.

10. The system of claim 6 , further comprising the model generator to:
determine a weight value of each of the plurality of generated failure models;
determine that the weight value of a generated failure model is below a threshold weight value; and
discard the generated failure model based on the determination that the weight value is below the threshold weight value.

11. A method, comprising:
extracting data from a plurality of sensors in a system monitoring a component, wherein the component is one of an electrical component and a mechanical component;
generating a plurality of features based on the extracted data;
analyzing a relevance of each of the plurality of generated features;
selecting a subset of the plurality of features;
determining that a feature of the plurality of generated features be used to generate failure models of the component based on a correlation to a failure type, wherein features with a first correlation above a threshold correlation are utilized to generate the failure models and features with a second correlation below the threshold correlation are discarded;
generating a plurality of models to model the first failure type of the component;
assembling the plurality of models into a single model for predicting a failure of the first failure type of the component;
extracting data associated with the failure predicted by the single model;
correlating data associated with a detected failure with the single model; and
determining when the plurality of features has successfully predicted the failure of the first failure type of the component.

12. The method of claim 11 , further comprising evaluating the health report for an accuracy of predicted component failures.

13. The method of claim 12 , wherein evaluating the health report includes:
extracting data associated with component failures; and
correlating the data associated with component failures with the health report.

14. The method of claim 11 , wherein generating the health report includes generating a predicted type of failure for a particular component.

15. The method of claim 11 , wherein analyzing the relevance of a feature comprises:
determining whether the feature predicts a selected failure; and
calculating the correlation of the feature to the prediction of the selected failure.

Assignees

• Hewlett Packard Development Co

Inventors

• Lopez Marlon Gomes
• Riberio Guilherme Selau
• Alves Luciano
• Marchezi Humberto Cardoso
• Tani Brian Schmitz
• Marquezini Leonardo
• Rodel Eduardo
• Gerent Josepe Franco

Classifications

• G05B2219/32201

  Build statistical model of past normal proces, compare with actual process · CPC title

• G05B23/024

  Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks · CPC title

• G06F11/3075

  the data filtering being achieved in order to maintain consistency among the monitored data, e.g. ensuring that the monitored data belong to the same timeframe, to the same system or component · CPC title

• G06N20/00

  Machine learning · CPC title

• G05B23/0243

  model based detection method, e.g. first-principles knowledge model · CPC title