Automatically adapting a prognostic-surveillance system to account for age-related changes in monitored assets

1. A method for automatically adapting a prognostic-surveillance system to account for aging phenomena in a monitored system, the prognostic-surveillance system comprising one or more machine-learning models, the method comprising:

receiving time-series signals associated with measurements obtained at one or more sensors in the monitored system;

analyzing, using a trained inferential machine-learning model, the time-series signals to detect incipient anomalies associated with the monitored system;

periodically determining a reward/cost metric associated with using an additional trained inferential machine-learning model trained to account for aging phenomena in the monitored system;

responsive to determining that the reward/cost metric exceeds a threshold, using the additional trained inferential machine-learning model to account for one or more aging phenomena in the monitored system;

using the additional trained inferential machine-learning model in the prognostic-surveillance system;

detecting, by the prognostic-surveillance system and using the additional trained inferential learning model, an incipient anomaly in the monitored system; and

in response to the detecting, performing a servicing operation on the monitored system to remediate the incipient anomaly.

2. The method of claim 1 , wherein the method further comprises:

using the trained inferential machine-learning model to generate estimated values for the time-series signals from the monitored system based on cross-correlations between the time-series signals;

performing pairwise differencing operations between actual values and the estimated values for the time-series signals set to produce residuals; and

analyzing the residuals to detect the incipient anomalies in the monitored system.

3. The method of claim 2 , wherein analyzing the residuals involves:

performing a sequential probability ratio test (SPRT) on the residuals to produce SPRT alarms; and

detecting the incipient anomalies based on the SPRT alarms.

4. The method of claim 1 , wherein the reward/cost metric balances an advantage gained by swapping the trained inferential machine-learning model against a cost associated with the swapping.

5. The method of claim 4 ,

wherein the advantage gained by swapping the trained inferential machine-learning model includes: reducing a number of false alarms that can possibly take the monitored system out of service; and increasing a sensitivity of the prognostic-surveillance system associated with detecting new degradation modes; and

wherein the cost associated with the swapping the trained inferential machine-learning model with the additional trained inferential machine-learning model includes model-retraining costs and model-swapping costs.

6. The method of claim 1 , wherein the method further comprises:

running the monitored system in a programmable environmental testing chamber to produce training data; and

using the training data to parametrically train the inferential machine-learning model across a range of ambient conditions with input from a subject matter expert, so that the trained inferential machine-learning model is able to discriminate between normal aging phenomena and degradation modes that lead to system failure.

7. The method of claim 1 , wherein the method further comprises training a set of age-specific machine-learning inferential models for the monitored system using age-specific data historian files, which contain time-series signals for similar monitored systems during different stages in life-cycles of the similar monitored systems, wherein each trained age-specific inferential machine-learning model is associated with different age-specific operational characteristics of the monitored system.

8. The method of claim 1 , wherein performing the servicing operation comprises one or more of the following:

performing an airflow-reversal operation to remove accumulated dust from components in a coal-fired power plant;

replacing one or more blades in a metal stamping press; and

replacing cooling fans in an enterprise computing system.

9. The method of claim 1 , wherein periodically determining the reward/cost metric and selectively swapping the trained inferential machine-learning model are performed by an insertable software module that can be easily inserted into an existing ML framework without having to replace an entire ML framework.

10. A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform operations for automatically adapting a prognostic-surveillance system to account for aging phenomena in a monitored system, the prognostic-surveillance system comprising one or more machine-learning models, the operations comprising:

receiving time-series signals associated with measurements obtained at one or more sensors in the monitored system;

analyzing, using a trained inferential machine-learning model, the time-series signals to detect incipient anomalies associated with the monitored system;

periodically determining a reward/cost metric associated with using an additional trained inferential machine-learning model trained to account for aging phenomena in the monitored system;

responsive to determining that the reward/cost metric exceeds a threshold, using with the additional trained inferential machine-learning model to account for one or more aging phenomena in the monitored system;

using the additional trained inferential machine-learning model in the prognostic-surveillance system;

detecting, by the prognostic-surveillance system and using the additional trained inferential learning model, an incipient anomaly in the monitored system; and

in response to the detecting, performing a servicing operation on the monitored system to remediate the incipient anomaly.

11. The non-transitory computer-readable storage medium of claim 10 , wherein the operations further comprises:

using the trained inferential machine-learning model to generate estimated values for the time-series signals from the monitored system based on cross-correlations between the time-series signals;

performing pairwise differencing operations between actual values and the estimated values for the time-series signals set to produce residuals; and

analyzing the residuals to detect the incipient anomalies in the monitored system.

12. The non-transitory computer-readable storage medium of claim 11 , wherein analyzing the residuals involves:

performing a sequential probability ratio test (SPRT) on the residuals to produce SPRT alarms; and

detecting the incipient anomalies based on the SPRT alarms.

13. The non-transitory computer-readable storage medium of claim 10 , wherein the reward/cost metric balances an advantage gained by swapping the trained inferential machine-learning model against a cost associated with the swapping.

14. The non-transitory computer-readable storage medium of claim 13 ,

wherein the advantage gained by swapping the trained inferential machine-learning model includes: reducing a number of false alarms that can possibly take the monitored system out of service; and increasing a sensitivity of the prognostic-surveillance system associated with detecting new degradation modes; and

wherein the cost associated with the swapping includes model-retraining costs and model-swapping costs.

15. The non-transitory computer-readable storage medium of claim 10 , wherein the operations further comprise:

running the monitored system in a programmable environmental testing chamber to produce training data; and

using the training data to parametrically train the inferential machine-learning model across a range of ambient conditions with input from a subject matter expert, so that the trained inferential machine-learning model is able to discriminate between normal aging phenomena and degradation modes that lead to system failure.

16. The non-transitory computer-readable storage medium of claim 10 , wherein the operations further comprise training a set of age-specific machine-learning inferential models for the monitored system using age-specific data historian files, which contain time-series signals for similar monitored systems during different stages in life-cycles of the similar monitored systems, wherein each trained age-specific inferential model is associated with different age-specific operational characteristics of the monitored system.

17. The non-transitory computer-readable storage medium of claim 10 , the operations further comprising:

responsive to detecting the incipient anomaly in the monitored system, performing a servicing operation on the monitored system to remediate the anomaly.

18. A prognostic-surveillance system for monitoring a monitored system, the prognostic-surveillance system comprising:

a non-transitory computer-readable medium storing computer-executable program instructions; and

a processing device communicatively coupled to the non-transitory computer-readable medium for executing the computer-executable program instructions, wherein executing the computer-executable program instructions configures the processing device to perform operations comprising:

receiving time-series signals associated with measurements obtained at one or more sensors in the monitored system;

analyzing, using a trained inferential machine-learning model, the time-series signals to detect incipient anomalies associated with the monitored system;

periodically determining a reward/cost metric associated with using an additional trained inferential machine-learning model trained to account for aging phenomena in the monitored system;

responsive to determining that the reward/cost metric exceeds a threshold, using the additional trained inferential machine-learning model to account for one or more aging phenomena in the monitored system;

using the additional trained inferential machine-learning model in the prognostic-surveillance system;

detecting, by the prognostic-surveillance system and using the additional trained inferential learning model, an incipient anomaly in the monitored system; and

in response to the detecting, performing a servicing operation on the monitored system to remediate the incipient anomaly.

19. The prognostic-surveillance system of claim 18 , wherein the reward/cost metric balances an advantage gained by swapping the trained inferential machine-learning model against a cost associated with the swapping.