Fault detection using event-based predictive models

Patent metadata

Field	Value
Publication number	US-10444121-B2
Application number	US-201615144846-A
Country	US
Kind code	B2
Filing date	May 3, 2016
Priority date	May 3, 2016
Publication date	Oct 15, 2019
Grant date	Oct 15, 2019

Abstract

A proposed solution to analyze event data from sensors of devices of a manufacturing environment to identify relationships of events to predict faults in devices is disclosed. The analysis includes conditional probability model and Apriori model. The relationships are used to determine a device health index which is compared to real-time event data to predict faults in a device.

First claim

The invention claimed is:

1. A method for diagnosing faults to predict faults comprising:
providing, to an analyzer server, event data of alarm events from sensors, comprising a first sensor and a second sensor, for measuring operating metrics of devices of a system of devices, wherein the event data comprises alarm events for a first device attached to the first sensor and a second device attached to the second sensor;
analyzing, by the analyzer server, the event data to identify relationships of the alarm events, wherein the analyzing comprises performing a conditional probability analysis to determine a causal relationship between the first sensor attached to the first device, and the second sensor attached to the second device, wherein the causal relationship indicates that a failure of the first device indicated by one of the alarm events that is an alarm event for the first sensor may lead to a failure of the second device;
defining, by the analyzer server, a device health index (DHI) based on the identified relationships of the alarm events;
monitoring, by the analyzer server, additional alarm events from the first sensor; and
predicting, by the analyzer server, a failure of the second device based on the monitored additional alarm events from the first sensor, the DHI, and the causal relationship between the first sensor and the second sensor.

2. The method of claim 1 wherein the event data comprises “Internet of Things” (IoT) data.

3. The method of claim 1 wherein the sensors comprise different types of sensors for sensing different operating metrics of devices.

4. The method of claim 1 , further comprising storing the event data in an event sensor log.

5. The method of claim 1 wherein fields of the event data comprise:
a sensor ID of a sensor;
a time stamp of a time when an alarm event occurred;
a message code indicating a type of alarm event; and
an input from an operator.

6. The method of claim 5 wherein the message code of the event data comprises one of:
abnormal sensor reading;
recovery from abnormal status;
threshold value changes; and
operator's input message.

7. The method of claim 1 , further comprising assigning an event ID to each event data based on a sequence of occurrence.

8. The method of claim 7 , further comprising categorizing the alarm events into a plurality of event types based on a sensor ID identifying a sensor of the alarm event and a message code indicating a type of event, wherein alarm events having a same sensor ID and message code are categorized as a same event type.

9. The method of claim 1 wherein performing the conditional probability analysis excludes event data which do not indicate a failure or a malfunction.

10. The method of claim 9 wherein for a given first event type (Event A) and a given second event type B (Event B) and if Event B always follows Event A within a defined period of time T, a probability of Event B occurring after Event A is determined by

P

A
,
B

T

:=

N
B

⁡

(

A
,
T

)

N
A

where
P A,B T is a probability that event B occurs after Event A within T,
N B (A,T) is a number of times that Event B occurred within T after an occurrence of Event A, and
N A is a number of times that Event A occurred.

11. The method of claim 10 wherein determining N B (A,T) comprises:
determining event pairs having Event B occurring within T after Event A, wherein an event pair includes event ID of Event A (Event A ID) and event ID of Event B (Event B ID);
assigning an event pair ID for an event pair;
listing event pairs in an order in a list;
initializing
N=1,
first event pair of the list=Last_Pair, and
next event pair of the list=Cur_Pair;

determining whether Cur_Pair is greater than Last_Pair,
if Cur_Pair is greater than Last_Pair, then
N=N+1, and

Last_Pair=Cur_Pair; and
determining whether all event pairs in the list have been analyzed,
if there are event pairs which have not been analyzed, then next event pair of the list=Cur_Pair,
and continue analysis from determining whether Cur_Pair is greater than Last_Pair, and

if there all event pairs in the list have been analyzed, then return N which is equal to NB(A,T).

12. The method of claim 8 wherein analyzing the event data comprises performing an Apriori analysis which mines the event data to identify frequent patterns.

13. The method of claim 12 wherein performing the Apriori analysis comprises:
transforming the event data into transactions; and
analyzing the transactions to identify frequent patterns to determine relationship between two transactions.

14. The method of claim 1 wherein the DHI comprises a 3-dimensional tuple, comprising:
an original reading from a sensor, of the sensors;
an average increment over a time window of readings from the sensor; and
a standard deviation of the readings from the sensor in the time window.

15. A system for diagnosing faults to predict faults comprising:
a database module, wherein the database module stores event data of events from sensors for measuring operating metrics of devices of a system of devices, wherein the event data comprises alarm events of devices; and
a processor module, wherein the processor module:
analyzes the event data to identify relationships of the alarm events, wherein the analyzing comprises performing a conditional probability analysis to determine a causal relationship between a first sensor, of the sensors, attached to a first device, of the devices, and a second sensor, of the sensors, attached to a second device, of the devices, wherein the causal relationship indicates that a failure of the first device indicated by an alarm event of the first sensor may lead to a failure of the second device,
monitors additional alarm events from the first sensor and compares the additional alarm events with a DHI which is defined based on the identified relationships of the alarm events, including the causal relationship between the first sensor and the second sensor, and
predicts a failure of the second device based on the monitored additional alarm events from the first sensor, the DHI and the causal relationship between the first sensor and the second sensor.

16. The system of claim 15 wherein the processor module comprises a stream event processor which monitors events from sensors and analyzes a stream data to predict failures based on the DHI.

17. The system of claim 15 wherein analyzing the event data further comprises performing an Apriori analysis which mines the event data to identify frequent patterns to determine relationships of events.

18. A non-transitory computer-readable medium having stored thereon program code, the program code executable by a computer, the program code comprising:
providing, to an analyzer server, event data of events from sensors for measuring operating metrics of devices of a system

Assignees

• Sap Se

Inventors

• Wang Chen

Classifications

• G01M99/008 Primary

  by doing functionality tests · CPC title

• G01M99/005

  Testing of complete machines, e.g. washing-machines or mobile phones (testing of machine parts G01M13/00; testing of electric apparatus or components G01R31/50) · CPC title