System and methods for automated plant asset failure detection

• US 10,192,170 B2
• Filed: 11/30/2016
• Issued: 01/29/2019
• Est. Priority Date: 03/15/2013
• Status: Active Grant

• Current Assignee: Aspentech Corporation
• Original Assignee: Mtelligence Corporation (Aspentech Corporation)
• Inventors: Bates, Alexander B., Rahilly, Paul, Macnab, Scott
• Primary Examiner(s): Ko, Chae M
• Application Number:
  US15/365,607
• Publication Number:
  US 20170083830A1
• Time in Patent Office: 790 Days
• Field of Search: 714 26, 714 473
• CPC Class Codes: G05B 23/0229 knowledge based, e.g. expert systems; genetic algorithms
• CPC Class Codes: G05B 23/024 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 Class Codes: G06F 11/0709 in a distributed system consisting of a plurality of standalone computer nodes, e.g. clusters, client-server systems
• CPC Class Codes: G06F 11/2257 using expert systems
• CPC Class Codes: G06F 21/552 involving long-term monitoring or reporting
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A system for performing failure signature recognition training for at least one unit of equipment. The system includes a memory and a processor coupled to the memory. The processor is configured by computer code to receive sensor data relating to the unit of equipment and to receive failure information relating to equipment failures. The processor is further configured to analyze the sensor data in view of the failure information in order to develop at least one learning agent for performing failure signature recognition with respect to the at least one unit of equipment.

Claims

1. A computer-implemented method of performing failure signature recognition training, the method comprising:

by one or more processors and associated memory;

receiving;

(i) sensor data relating to at least one unit of equipment and (ii) failure information relating to equipment failures;

analyzing the received sensor data in view of the received failure information, the analyzing being automatic in response to the sensor data and;

creating at least one learning agent to perform failure signature recognition with respect to the at least one unit of equipment; and

training the at least one learning agent by adjusting parameters of the at least one learning agent using machine learning, the training enabling the at least one learning agent to predict failures identified by the received failure information; and

storing the received sensor data in the associated memory along with metadata flagging one or more failure intervals and one or more normal intervals in the received sensor data.

2. The method of claim 1, wherein the received failure information relates to prior failures of the at least one unit of equipment.

3. The method of claim 1, wherein the received failure information relates to prior failures of different equipment than the at least one unit of equipment, and wherein the different equipment is functionally similar to the at least one unit of equipment.

4. The method of claim 1, wherein the received failure information includes indications of the equipment failures, the indications containing at least one of:

indication types, dates, and codes associated with the equipment failures.

5. The method of claim 1, further comprising:

receiving an input identifying one of;

(a) a group of equipment including the at least one unit of equipment or (b) only the at least one unit of equipment.

6. The method of claim 5, wherein the input identifies a plurality of tags corresponding to a plurality of sensors associated with the at least one unit of equipment.

7. The method of claim 6, wherein the input further identifies at least one of:

(1) minimum outlier removal settings for one or more of the plurality of sensors and (2) maximum outlier removal settings for one or more of the plurality of sensors.

8. The method of claim 1, wherein further training the at least one learning agent by biasing the parameters to avoid at least one of:

false positive errors and false negative errors.

9. The method of claim 1, wherein the training further enables the at least one learning agent to achieve a desired potential failure to failure (P-F) interval.

10. The method of claim 9, wherein the desired P-F interval is a widest P-F interval obtainable in view of a desired level of failure prediction accuracy.

11. The method of claim 1, wherein the failure information comprises work order information, and the method further comprising:

identifying faults in the work order information that correspond to the equipment failures; and

classifying the identified faults to predict failures by the trained at least one learning agent.

12. The method of claim 1, further comprising:

analyzing the received sensor data at time of one or more of the equipment failures in order to learn one or more fault signatures, wherein the one or more fault signatures are one of;

a memory signature or a non-memory signature.

13. The method of claim 1, further comprising:

receiving (i) additional sensor data relating to an additional unit of equipment and (ii) additional failure information relating to one or more additional equipment failures;

analyzing the additional sensor data in view of the one or more additional equipment failures to learn one or more fault signatures; and

storing the learned one or more fault signatures in the associated memory.

14. The method of claim 1, further comprising:

tuning a prediction interval over which the received sensor data is evaluated, the tuning evaluating failure prediction accuracy over multiple interval durations.

15. A computer-implemented method of performing anomaly detection, the method comprising:

by one or more processors and associated memory;

receiving;

(i) sensor data relating to at least one unit of equipment and (ii) failure information relating to one or more equipment failures;

automatically determining one or more normal operating states of the at least one unit of equipment, the determining performed by analyzing the received sensor data over time periods different than periods of the one or more equipment failures; and

training at least one anomaly agent to detect when a current operating state of the at least one unit of equipment is outside the determined one or more normal operating states, wherein the training is automatically performed by the one or more processors and includes;

modeling the determined one or more normal operating states using a training model;

monitoring for additional sensor data relating to the at least one unit of equipment;

providing the additional sensor data to the training model, the training model determining an error associated with classifying the additional sensor data into an operating state of the determined one or more normal operating states;

checking whether the determined error meets an anomaly threshold; and

detecting an anomaly condition based upon results of the checking.

16. The method of claim 15, wherein the training model is a self-organizing map, and the training further comprises:

modeling the determined one or more normal operating states using the self-organizing map;

providing the additional sensor data to the self-organizing map, the self-organizing map determining the error associated with classifying the additional sensor data into the operating state of the determined one or more normal operating states;

comparing the determined error to a maximum error to check if the determined error meets the anomaly threshold; and

detecting the anomaly condition based upon the comparison.

17. The method of claim 15, wherein the training model is a Restricted Boltzman Machine (RBM).

18. The method of claim 15, wherein modeling the determined one or more normal operating states is based on Gaussian probabilities.

19. The method of claim 15, wherein the at least one anomaly agent is configured to model the determined one or more normal operating states and monitor for the additional sensor data.

20. The method of claim 15, further comprising:

creating a supervised learning profile agent configured to learn details of the detected anomaly condition.