Oil debris monitoring (ODM) with adaptive learning

• US 10,409,275 B2
• Filed: 10/19/2016
• Issued: 09/10/2019
• Est. Priority Date: 10/19/2016
• Status: Active Grant

• Current Assignee: Rtx Corporation
• Original Assignee: United Technologies Corporation (Raytheon Technologies Corporation d/b/a RTX)
• Inventors: Hagen, Gregory S., Lin, Yiqing, Erdinc, Ozgur, Giering, Michael J., Khibnik, Alexander I.
• Primary Examiner(s): Dao, Tuan C
• Application Number:
  US15/297,319
• Publication Number:
  US 20180107203A1
• Time in Patent Office: 1,056 Days
• Field of Search: None
• CPC Class Codes: F01D 21/003 Arrangements for testing or measuring for measuring vibrations G01H
• CPC Class Codes: F01D 25/16 Arrangement of bearings; Supporting or mounting bearings in casings bearings per se F16C
• CPC Class Codes: F01D 25/18 Lubricating arrangements of machines or engines in general F01M
• CPC Class Codes: F05D 2220/32 in gas turbines
• CPC Class Codes: F16H 57/0405 Monitoring quality of lubricant or hydraulic fluids
• CPC Class Codes: F16N 29/00 Special means in lubricating arrangements or systems providing for the indication or detection of undesired conditions; Use of devices responsive to conditions in lubricating arrangements or systems
• CPC Class Codes: G01N 33/2835 specific substances contained in the oil or fuel
• CPC Class Codes: G01N 33/2858 metal particles
• CPC Class Codes: G05B 13/0265 the criterion being a learning criterion
• CPC Class Codes: G05B 23/0205 by means of a monitoring system capable of detecting and responding to faults
• CPC Class Codes: G05B 23/0289 Reconfiguration to prevent failure, e.g. usually as a reaction to incipient failure detection
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: Y02E 10/72 Wind turbines with rotation axis in wind direction

Abstract

A system and method for debris particle detection with adaptive learning are provided. The method includes receiving oil debris monitoring (ODM) sensor data from an oil debris monitor sensor and fleet data from a database, detecting a feature in the ODM sensor data, generating an anomaly detection signal based on detecting an anomaly by comparing the feature in the ODM sensor data to a limit defined by system information stored in the fleet data, selecting a maintenance action request based on the anomaly detection signal, and adjusting one or more of the feature, the anomaly, the limit, and the maintenance action request by applying an adaptive learning algorithm that uses the ODM sensor data, fleet data, and feedback from field maintenance of one or more engines that evolves over time.

Claims

1. A method for debris particle detection with adaptive learning in a gas turbine engine, the method comprising:

delivering oil through a mechanical system including at least one of an active flow valve and a passive flow valve;

sensing a flow of the oil using an oil debris monitor sensor;

receiving, at a signal processor, oil debris monitoring (ODM) sensor data from the oil debris monitor sensor based on the sensed flow of the oil, and received, at the signal processor, fleet data from a database;

detecting, via the signal processor, a feature in the ODM sensor data;

generating, via the signal processor, an anomaly detection signal based on detecting an anomaly by comparing the feature in the ODM sensor data to a limit defined by system information stored in the fleet data;

selecting a maintenance action request based on the anomaly detection signal;

adjusting one or more of the feature, the anomaly, the limit, and the maintenance action request by applying, via an adaptive learning algorithm implementing processor, an adaptive learning algorithm that uses the ODM sensor data, fleet data, and feedback from field maintenance of one or more engines that evolves over time to improve an accuracy of the adaptive learning algorithm to detect the feature in the ODM sensor data; and

applying the adaptive learning algorithm to on-board parameters to detect particles in real-time,

wherein applying adaptive learning on ODM sensor data further comprises training a first a set of historical sensor data from fleet data to differentiate the characteristics of parameters with or without a debris particle.

2. The method of claim 1, wherein the adaptive learning algorithm is one selected from a group consisting of a machine learning algorithm, a support vector machine (SVM) algorithm, and a supervised machine learning classification algorithm.

3. The method of claim 1, further comprising:

collecting field maintenance data to serve as ground truth of particle detection.

4. The method of claim 3, wherein field maintenance data includes one or more of data indicating whether a real particle is detected, data indicating signal noise has been falsely identified as particles, and data indicating sensor system fault, wherein as the adaptive learning algorithm continues to learn from the field maintenance data, a success rate at which the real particle is detected is increased and a false alarm rate is improved.

5. The method of claim 1, further comprising a fine-tuning procedure that includes:

finding a discrepancy between a detection algorithm that is used to detect the features in the ODM sensor data and fleet data;

generating additional training data from the discrepancy; and

adding the additional training data to a training set to fine-tune the adaptive learning algorithm.

6. The method of claim 5, further comprising:

applying the fine-tuning procedure to data from at least one selected from a group including a single engine, a collection of engines in the fleet data, and engines from a different fleet.

7. The method of claim 5, further comprising:

applying the fine-tuning procedure to one or more selected from independently to each individual engine and a group of engines to reflect its unique characteristics.

8. The method of claim 1, wherein applying the adaptive learning algorithm continues until a detection algorithm produces detection accuracy that meets or exceeds a detection accuracy threshold.

9. The method of claim 1, further comprising:

collecting field maintenance data from a subset of engine data obtained from a limited set stored in fleet data.

10. The method of claim 1, further comprising:

adjusting the adaptive particle detection algorithm if additional ODM sensor parameters are included in fleet data at any time; and

re-training adaptive particle detection algorithms.

11. The method of claim 1, further comprising:

adjusting one or more of the feature, the anomaly, the limit, and the maintenance action request upon receiving feedback from field maintenance of one or more engines.

12. The method of claim 1, further comprising:

enabling particle detection algorithms to learn from actual data from field maintenance; and

adjusting a discrepancy learned from field maintenance.

13. A system for debris particle detection with adaptive learning in a gas turbine engine, the system comprising:

a memory having computer readable instructions; and

a processor configured to execute the computer readable instructions, the computer readable instructions comprising;

receiving oil debris monitoring (ODM) sensor data from an oil debris monitor sensor installed in a mechanical system configured to flow oil through including at least one of an active flow valve and a passive flow valve, the oil debris monitor sensor configured to sense the flow of the oil;

receiving fleet data from a database;

detecting a feature in the ODM sensor data;

generating an anomaly detection signal based on detecting an anomaly by comparing the feature in the ODM sensor data to a limit defined by system information stored in the fleet data;

selecting a maintenance action based on the anomaly detection signal;

adjusting one or more of the feature, the anomaly, the limit, and the maintenance action request by applying an adaptive learning algorithm that uses the ODM sensor data, fleet data, and feedback from field maintenance of one or more engines that evolves over time to improve an accuracy of the adaptive learning algorithm to detect the feature in the ODM sensor data; and

applying the adaptive learning algorithm to on-board parameters to detect particles in real-time,

wherein applying adaptive learning on ODM sensor data further comprises training a first a set of historical sensor data from fleet data to differentiate the characteristics of parameters with or without a debris particle.

14. The system of claim 13, wherein the computer readable instructions further comprises:

collecting field maintenance data to serve as ground truth of particle detection.

15. The system of claim 13, wherein the computer readable instructions further comprises a fine-tuning procedure that includes:

finding a discrepancy between a detection algorithm that is used to detect the features in the ODM sensor data and fleet data;

generating additional training data from the discrepancy; and

adding the additional training data related to the discrepancy to a training set to fine-tune the adaptive learning algorithm.

16. A computer program product for debris particle detection with adaptive learning, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:

receive oil debris monitoring (ODM) sensor data from an oil debris monitor sensor and fleet data from a database;

detect, via a signal processor, a feature in the ODM sensor data;

generate, via the signal processor, an anomaly detection signal based on detecting an anomaly by comparing the feature in the ODM sensor data to a limit defined by system information stored in the fleet data;

select a maintenance action request based on the anomaly detection signal;

adjust one or more of the feature, the anomaly, the limit, and the maintenance action request by applying, via an adaptive learning algorithm implementing processor, an adaptive learning algorithm that uses the ODM sensor data, fleet data, and feedback from field maintenance of one or more engines that evolves over time to improve an accuracy of the adaptive learning algorithm to detect the feature in the ODM sensor data; and

applying the adaptive learning algorithm to on-board parameters to detect particles in real-time,

wherein applying adaptive learning on ODM sensor data further comprises training a first a set of historical sensor data from fleet data to differentiate the characteristics of parameters with or without a debris particle.