Sensor time series data: functional segmentation for effective machine learning

Patent metadata

Field	Value
Publication number	US-10936600-B2
Application number	US-201615331258-A
Country	US
Kind code	B2
Filing date	Oct 21, 2016
Priority date	Oct 23, 2015
Publication date	Mar 2, 2021
Grant date	Mar 2, 2021

Abstract

Feature engineering can be performed on time series data making the data easy to manipulate and accessible to business users for analysis according to existing best practices. A computer system can, after receiving time series data related to a device, contextualize the time series data based on business data related to the device from, for example, an enterprise resource planning database. The contextualized data can be windowed by a selected feature based on execution data related to the device from, for example, a manufacturing execution system database. The windowed data can be transformed into summary data using a time series transformation. The summary data can be easily manipulated by, for example, generating genetic maps of the segmented and transformed data for clustering or searching for anomalies and patterns in response to user requests or automatically.

First claim

What is claimed is:

1. A method for performing feature engineering of time series data, the method comprising:
receiving, at a computer system, time series data collected from a sensor of a manufacturing machine for a period of time, the time series data providing information about the manufacturing machine for the period of time;
segmenting, by the computer system, the time series data based on business data related to the manufacturing machine for the period of time to generate contextualized data segments, the business data comprising operational data of the manufacturing machine and received from a business database;
selecting, automatically by the computer system without user input, a time window type based on the contextualized data segments;
feature windowing, by the computer system, the contextualized data segments into time windows of the time window type and based on execution data related to the manufacturing machine for the period of time to generate a plurality of windows of data, the execution data associated with execution of the manufacturing machine and received from a manufacturing database;
transforming, by the computer system, each of the plurality of windows of data into summary data using a time series transformation to generate a plurality of summary data,

wherein transforming comprises:
generating a plurality of strings corresponding to the plurality of windows of data;
converting the plurality of strings into a corresponding plurality of genetic maps;
clustering the plurality of genetic maps to identify pattern similarities; and
grouping the plurality of genetic maps based on the pattern similarities; and
storing, by the computer system, each of the plurality of summary data.

2. The method for performing feature engineering of time series data of claim 1 , wherein generating the plurality of strings corresponding to the plurality of windows of data comprises, for each window of data in the plurality of windows of data:
normalizing the data in the window of data;
smoothing the data in the window of data using piecewise aggregate approximation; and
converting the data in the window of data into a string.

3. The method for performing feature engineering of time series data of claim 1 , the method further comprising:
displaying groupings of the plurality of genetic maps to a user.

4. The method for performing feature engineering of time series data of claim 1 , the method further comprising:
identifying, by the computer system, a pattern that represents an indication of a device failure;
building, by the computer system, a predictive model based on the pattern; and
sending, by the computer system, an alert based on applying the predictive model to current data.

5. The method for performing feature engineering of time series data of claim 1 , the method further comprising:
receiving, by the computer system, a selection of a second time series data from a user, the second time series data related to the manufacturing machine for a second period of time;
transforming, by the computer system, the second time series data into selected summary data using the time series transformation;
searching, by the computer system, the plurality of summary data for a pattern matching at least a portion of the selected summary data; and
displaying, by the computer system, summary data from the plurality of summary data having the pattern to the user.

6. The method for performing feature engineering of time series data of claim 5 , the method further comprising:
predicting, by the computer system, a quality of a second product produced by the manufacturing machine during the second period of time based on a quality of a first product produced by the manufacturing machine during the period of time having the matching pattern.

7. The method for performing feature engineering of time series data of claim 1 , wherein the time series transformation used is symbolic aggregate approximation (“SAX”) and each of the plurality of summary data is a string in a plurality of strings.

8. The method for performing feature engineering of time series data of claim 1 , wherein the time window type is one of a time segment, a sliding window, or an event.

9. The method for performing feature engineering of time series data of claim 1 , wherein the business data is data from an enterprise resource planning database and the execution data is from a manufacturing execution system database.

10. A system for performing feature engineering of time series data, the system comprising:
one or more processors; and
one or more memories having stored thereon instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
receiving time series data collected from a sensor of a manufacturing machine for a period of time, the time series data providing information about the manufacturing machine for the period of time;
segmenting the time series data based on business data related to the manufacturing machine for the period of time to generate contextualized data segments, the business data comprising operational data of the manufacturing machine and received from a business database;
selecting, automatically without user input, a time window type based on the contextualized data segments;
feature windowing the contextualized data segments into time windows of the time window type and based on execution data related to the manufacturing machine for the period of time to generate a plurality of windows of data, the execution data associated with execution of the manufacturing machine and received from a manufacturing database;
transforming each of the plurality of windows of data into summary data using a time series transformation to generate a plurality of summary data, wherein transforming comprises:
generating a plurality of strings corresponding to the plurality of windows of data;
converting the plurality of strings into a corresponding plurality of genetic maps;
clustering the plurality of genetic maps to identify pattern similarities; and
grouping the plurality of genetic maps based on the pattern similarities; and
storing each of the plurality of summary data.

11. The system for performing feature engineering of time series data of claim 10 , wherein generating the plurality of strings corresponding to the plurality of windows of data comprises, for each window of data in the plurality of windows of data:
normalizing the data in the window of data;
smoothing the data in the window of data using piecewise aggregate approximation; and
converting the data in the window of data into a string.

12. The system for performing feature engineering of time series data of claim 11 , wherein the operations further comprise:
displaying groupings of the plurality of genetic maps to a user.

13. The system for performing feature engineering of time series data of claim 10 , wherein the operations further comprise:
identifying a pattern that represents an indication of a device failure;
building a predictive model based on the pattern; and
sending an alert based on applying the predictive model to current data.

14. The system for performing feature engineering of time series data of claim 10 , wherein the operations further comprise:
receiving a selection of a second time series data from a user, the second time series data related to the manufacturing machine for a second period of time;
transforming the second time series data into selected summary data using the time series transformation;
searching the plurality of summary data for a pat

Assignees

• Oracle Int Corp

Inventors

• Hariharan Sreeram
• Raman Ramchand
• Ratnam Gopal Raja
• So Bryan Siu Him
• Arora Usha
• Nookala Ganga Mohan
• Jonnalagadda Krishna
• Nagasuri Pushkala
• Uppala Swathi
• Mani Sangeetha

Classifications

• G06N20/00

  Machine learning · CPC title

• G06F16/258

  Data format conversion from or to a database · CPC title

• G06F16/285

  Clustering or classification · CPC title

• G06F16/24575 Primary

  using context · CPC title

• G06F16/248

  Presentation of query results · CPC title