Apparatus and method for using model training and adaptation to detect furnace flooding or other conditions

Patent metadata

Field	Value
Publication number	US-10746405-B2
Application number	US-201715722600-A
Country	US
Kind code	B2
Filing date	Oct 2, 2017
Priority date	Apr 24, 2017
Publication date	Aug 18, 2020
Grant date	Aug 18, 2020

Abstract

A method includes obtaining data associated with operation of equipment in an industrial process and identifying training data and evaluation data in the obtained data. The method also includes, during each of multiple training periods, identifying one or more models and one or more first statistical values using at least some of the training data and determining a threshold value using the one or more first statistical values. The one or more models represent the operation of the equipment. The method further includes, during each of multiple evaluation periods, determining one or more second statistical values using at least some of the evaluation data and the one or more models, comparing the one or more second statistical values to the threshold value determined in a preceding one of the training periods, and determining whether the equipment is suffering from at least one specified condition based on the comparison. In addition, the method includes, in response to determining that the equipment is suffering from the at least one specified condition, generating an alert identifying the at least one specified condition.

First claim

What is claimed is:

1. A method for generating timely alerts to move equipment in an industrial process to a safe operating condition comprising:
obtaining data associated with an operation of equipment in an industrial process, wherein the data is obtained utilizing at least one sensor comprising a structure for measuring at least one characteristic in or associated with the equipment in the industrial process, the data including an operating condition of the equipment in the industrial process;
identifying training data and evaluation data in the obtained data;
during each of multiple training periods, identifying one or more models and one or more first statistical values using at least some of the training data and determining a threshold value using the one or more first statistical values, the one or more models representing the operation of the equipment;
during each of multiple evaluation periods, determining one or more second statistical values using at least some of the evaluation data and the one or more models, wherein determining the one or more second statistical values for one of the evaluation periods comprises, further comprises:
normalizing each of multiple process variable signals using signal parameters estimated during a preceding training period, the process variable signals identifying values of process variables associated with the equipment;
generating predictions for the process variable signals based on regression coefficients estimated during the preceding training period;
identifying errors between the predictions and the normalized process variable signals;
scaling the errors based on standard deviations of residuals estimated during the preceding training period;
for each of multiple time instants during the evaluation period, identifying a sum of squared scaled residuals; and
identifying multiple second statistical values for the multiple time instants using the sum of the squared scaled residuals,
comparing the one or more second statistical values to the threshold value determined in a preceding one of the training periods, and
determining whether the equipment is suffering from at least one specified condition based on the comparison; and

in response to determining that the equipment is suffering from the at least one specified condition, generating a timely alert identifying the at least one specified condition to move the equipment in an industrial process to a safe operating condition.

2. The method of claim 1 , wherein determining the one or more first statistical values for one of the training periods comprises:
estimating distribution parameters for multiple process variable signals, the process variable signals identifying values of process variables associated with the equipment;
normalizing the process variable signals using the distribution parameters;
performing multivariate regression for the normalized process variable signals to generate regression equations;
using regression coefficients of the regression equations, identifying standard deviations of resulting residuals for the regression equations;
scaling the residuals using the standard deviations;
for each of multiple time instants during the training period, identifying the sum of squared scaled residuals; and
identifying multiple first statistical values for the multiple time instants using the sum of squared scaled residuals.

3. The method of claim 2 , wherein determining the threshold value for one of the training periods comprises:
identifying distribution parameters, a mean, and a standard deviation of the multiple first statistical values generated for the training period; and
identifying the threshold value based on the distribution parameters, the mean, and the standard deviation of the multiple first statistical values.

4. The method of claim 1 , wherein the at least one sensor comprises at least one of:
an oxygen sensor that measures an oxygen level located at at least one location with the equipment;
a pressure sensor that measures a pressure level at the at least one location within the equipment;
a temperature sensor that measures a temperature at the at least one location within the equipment or which measures a temperature of a process fluid being heated by the equipment;
a sensor that measures an amount of combustible material at the at least one location within the sensors; and
a sensor that measures a flow of material into the equipment.

5. The method of claim 1 , wherein:
the evaluation periods occur at a first interval; and
the training periods occur at a second interval longer than the first interval and are performed to retrain the one or more models to changing conditions or operational modes of the equipment.

6. The method of claim 1 , wherein identifying the training data and the evaluation data comprises:
identifying data associated with a specified length of time preceding a current time;
discarding data in a most-recent segment of the specified length of time;
in a remaining portion of the specified length of time, discarding data associated with a segment of time when a process variable signal is outside of a specified range; and
using non-discarded data in the remaining portion of the specified length of time as the training data for a current training period.

7. The method of claim 6 , wherein identifying the training data and the evaluation data further comprises at least one of:
discarding a segment of the data immediately before and a segment of the data immediately after the segment of time when the process variable signal is outside of the specified range; and
discarding a segment of the data spanning less than a threshold amount of time located between two discarded segments of data.

8. The method of claim 1 , wherein:
obtaining the data comprises obtaining measurements from multiple sensors including the at least one sensor, wherein at least two of the multiple sensors capture measurements at different frequencies; and
the method further comprises synchronizing the measurements from the at least two of the multiple sensors to a common time scale.

9. The method of claim 1 , wherein:
the equipment in the industrial process comprises a furnace;
the obtained data comprises measurements from multiple sensors associated with the furnace; and
the at least one specified condition comprises flooding of the furnace.

10. The method of claim 9 , wherein obtaining the data, identifying the training data and the evaluation data, identifying the one or more models, determining whether the equipment is suffering from the at least one specified condition, and generating the alert are repeated for each of multiple furnaces.

11. An apparatus for generating timely alerts to move equipment in an industrial process to a safe operating condition comprising:
at least one processing device configured to:
obtain data associated with operation of equipment in an industrial process, wherein the data is obtained utilizing at least one sensor comprising a structure for measuring at least one characteristic in or associated with the equipment in the industrial process, the data including an operating condition of the equipment in the industrial process;
identify training data and evaluation data in the obtained data;
during each of multiple training periods, identify one or more models and one or more first statistical values using at least some of the training data and determine a threshold value using the one or more first statistical values, the one or more models representing the operation of the equipment;
during each of multiple evaluation periods, determine one or more second statistical values using at least some

Assignees

• Honeywell Int Inc

Inventors

• Vitt Constantine A
• Olecka Anna
• Miller Randy M
• Leclerc Francois
• Foslien Wendy K
• He Hao
• Groves William

Classifications

• F23N5/265

  using electronic means · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• F23N2005/185

  using detectors sensitive to rate of flow of fuel · CPC title

• F23N5/242

  using electronic means · CPC title

• F23N2225/08

  temperature · CPC title