Remote data analytics to predict system components or device failure

• US 10,139,788 B2
• Filed: 03/15/2016
• Issued: 11/27/2018
• Est. Priority Date: 03/15/2016
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Dutta, Manas, Shetty, Praveen, Koniki, Ramesh Babu, Gurrapu, Praveen, Gellaboina, Mahesh Kumar, Mallavarpu, Sreedhara
• Primary Examiner(s): Suryawanshi, Suresh
• Application Number:
  US15/071,104
• Publication Number:
  US 20170269562A1
• Time in Patent Office: 987 Days
• Field of Search: 700 29
• CPC Class Codes: G05B 13/048 using a predictor
• CPC Class Codes: G05B 23/0229 knowledge based, e.g. expert systems; genetic algorithms

Abstract

An apparatus includes a memory and one or more processors operably connected to the memory. The one or more processors are configured to receive data collected from a process facility system, detect anomalies for field device or process failures associated with the process facility system that are not monitored by alarms, detect leading indicators for field device or process failures that are monitored by alarms, and monitor the process facility system to detect further anomalies and leading indicators before failures occur.

Claims

1. An apparatus comprising:

a memory; and

one or more processors operably connected to the memory, the one or more processors configured to;

receive data collected from a field device or a process in a process facility system;

compute a pairwise correlation of parameters from the collected data indicating a strength and a direction of an interaction between each pair of parameters;

filter the parameters with the pairwise correlation having the strength and the direction outside of a range;

detect an anomaly based on the filtered parameters for an unsupervised failure of the field device or the process, wherein the unsupervised failure is a failure that is not monitored by alarms with an alarm;

detect a leading indicator based on the filtered parameters for a supervised failure of the field device or the process, wherein the supervised failure is a failure monitored with the alarm; and

monitor current data collected from the process facility system for the anomaly and the leading indicator before a future unsupervised failure or a future supervised failure occurs.

2. The apparatus of claim 1, wherein to detect the anomaly, the one or more processors are further configured to:

extract time series data from the collected data;

convert the collected data to a regularized time series;

apply a correlation to reduce dimensionality of the collected data converted to the regularized time series;

detect outlier data from the collected data; and

identify a distribution for normal behavior from the collected data after the outlier data is removed.

3. The apparatus of claim 2, wherein to monitor the current data collected for the anomaly before the future unsupervised failure occurs, the one or more processors are further configured to identify the anomaly from the normal behavior distribution.

4. The apparatus of claim 2, wherein:

the correlation comprises a Pearson correlation, and

a principal component analysis is used to detect the outlier data.

5. The apparatus of claim 1, wherein to detect the leading indicator, the one or more processors are further configured to:

identify a trigger from alarm data of the process facility system;

determine one or more correlations using parameters related to the identified trigger;

perform dimensionality reduction on the determined one or more correlations; and

create a predictive model for the identified trigger.

6. The apparatus of claim 5, wherein to monitor the current data collected for the leading indicator before the future supervised failure occurs, the one or more processors are further configured to deploy the leading indicator for the alarm from the predictive model.

7. The apparatus of claim 5, wherein a selection technique for performing dimensionality reduction is one of:

random forest or lasso.

8. A method comprising:

receiving data collected from a field device or a process in a process facility system;

computing a pairwise correlation of parameters from the collected data indicating a strength and a direction of an interaction between each pair of parameters;

filtering the parameters with the pairwise correlation having the strength and the direction outside of a range;

detecting an anomaly based on the filtered parameters for an unsupervised failure of the field device or the process, wherein the unsupervised failure is a failure that is not monitored with an alarm;

detecting a leading indicator based on the filtered parameters for a supervised failure of the field device or the process, wherein the supervised failure is a failure monitored with the alarm; and

monitoring current data collected from the process facility system for the anomaly and the leading indicator before a future unsupervised failure or a future supervised failure occurs.

9. The method of claim 8, wherein detecting the anomaly further comprises:

extracting time series data from the collected data;

converting the collected data to a regularized time series;

applying a correlation to reduce dimensionality of the collected data converted to the regularized time series;

detecting outlier data from the collected data; and

identifying a distribution for normal behavior from the collected data after the outlier data is removed.

10. The method of claim 9, wherein monitoring the current data collected for the anomaly before the future unsupervised failure occurs comprises identifying the anomaly from the normal behavior distribution.

11. The method of claim 9, wherein:

the correlation comprises a Pearson correlation, and

a principal component analysis is used to detect the outlier data.

12. The method of claim 8, wherein detecting the leading indicator comprises:

identifying a trigger from alarm data of the process facility system;

determining one or more correlations using parameters related to the identified trigger;

performing dimensionality reduction on the determined one or more correlations; and

creating a predictive model for the identified trigger.

13. The method of claim 12, wherein monitoring the current data collected for the leading indicator before the future supervised failure occurs comprises deploying the leading indicator for the alarm from the predictive model.

14. The method of claim 12, wherein a selection technique for performing dimensionality reduction is one of:

random forest or lasso.

15. A non-transitory machine-readable medium encoded with executable instructions that, when executed, cause one or more processors to:

receive data collected from a field device or a process in a process facility system;

compute a pairwise correlation of parameters from the collected data indicating a strength and a direction of an interaction between each pair of parameters;

filter the parameters with the pairwise correlation having the strength and the direction outside of a range;

detect an anomaly based on the filtered parameters for an unsupervised failure of the field device or the process, wherein the unsupervised failure is a failure that is not monitored with an alarm;

detect a leading indicator based on the filtered parameters for a supervised failure of the field device or the process, wherein the supervised failure is a failure monitored with the alarm; and

monitor current data collected from the process facility system for the anomaly and the leading indicator before a future unsupervised failure or a future supervised failure occurs.

16. The non-transitory machine-readable medium of claim 15, wherein the instructions to detect the anomaly further comprise instructions to:

extract time series data from the collected data;

convert the collected data to a regularized time series;

apply a correlation to reduce dimensionality of the collected data converted to the regularized time series;

detect outlier data from the collected data; and

identify a distribution for normal behavior from the collected data after the outlier data is removed.

17. The non-transitory machine-readable medium of claim 16, wherein the instructions to monitor the current data collected for the anomaly before the future unsupervised failure occurs comprise instructions to identify the anomaly from the normal behavior distribution.

18. The non-transitory machine-readable medium of claim 16, wherein:

the correlation comprises a Pearson correlation, and

a principal component analysis is used to detect the outlier data.

19. The non-transitory machine-readable medium of claim 15, wherein the instructions to detect the leading indicator comprise instructions to:

identify a trigger from alarm data of the process facility system;

determine one or more correlations using parameters related to the identified trigger;

perform dimensionality reduction on the determined one or more correlations; and

create a predictive model for the identified trigger.

20. The non-transitory machine-readable medium of claim 19, wherein the instructions to monitor the current data for the leading indicator before the future supervised failure occurs comprise instructions to deploy the leading indicator for the alarm from the predictive model.