FAULT DETECTION USING EVENT-BASED PREDICTIVE MODELS

• US 20170322120A1
• Filed: 05/03/2016
• Published: 11/09/2017
• Est. Priority Date: 05/03/2016
• Status: Active Grant

• Current Assignee: SAP SE
• Original Assignee: SAP SE
• Inventors: WANG, Chen
• Granted Patent: US 10,444,121 B2
• Time in Patent Office: Days
• CPC Class Codes: G01M 99/005 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 Class Codes: G01M 99/008 by doing functionality tests

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.

Claims

1. A method for diagnosing faults to predict faults comprising:

providing, to an analyzer server, 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;

analyzing the event data to identify relationships of events;

defining a device health index (DHI) based on identified relationships of events to predict failures based on the DHI; and

monitoring, by the analyzer server, events from sensors of the devices to determine possible failures based on the DHI.

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 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 the sensor;

a time stamp of the time when the alarm event occurred;

a message code indicating the 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 comprising assigning an event ID to each event data based on the sequence of occurrence.

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

9. The method of claim 8 wherein analyzing the alarm event data comprises performing a conditional probability analysis to determine causal relationship between any two sensors (sensor pair) of the sensors of the system of devices.

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

11. The method of claim 10 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, the probability of Event B occurring after Event A is determined by

12. The method of claim 11 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 N_B(A,T).

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

14. The method of claim 13 wherein performing the Apriori analysis comprises:

transforming event data into transactions; and

analyzing the transactions to identify frequent patterns to determine relationship between two transactions.

15. The method of claim 1 wherein the DHI comprises a 3-dimensional tuple which includes:

a first element of the tuple comprises normalized reading of first tuple a three dimensional tuple.

16. 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 events,

monitors events and comparing the events with a DHI which is defined based on identified relationships of events,

predicts failures based on the DHI.

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

18. The system of claim 16 wherein the processor module analyzes the event data using a conditional probability analysis to determine causal relationship between any two sensors (sensor pair) of the sensors of the system of devices.

19. The system of claim 16 wherein analyzing the event data comprises performing an Apriori analysis which mines the event data to identify frequent patterns to determine relationships of events.

20. 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 of devices, wherein the event data comprises alarm events of devices;

analyzing the event data to identify relationships of events using a conditional probability analysis and an Apriori analysis; and

monitoring, by the analyzer server, events from sensors of the devices to determine possible failures based on the DHI which is defined based on identified relationships of events to predict failures based on the DHI.