Methods and Systems for Discovery of Prognostic Subsequences in Time Series

Patent metadata

Field	Value
Publication number	US-2018164794-A1
Application number	US-201615375473-A
Country	US
Kind code	A1
Filing date	Dec 12, 2016
Priority date	Dec 12, 2016
Publication date	Jun 14, 2018
Grant date	—

Abstract

Systems and methods for determining a pattern in time series data representing an operation of a machine. A memory to store and provide a set of training data examples generated by a sensor of the machine, wherein each training data example represents an operation of the machine for a period of time ending with a failure of the machine. A processor configured to iteratively partition each training data example into a normal region and an abnormal region, determine a predictive pattern absent from the normal regions and present in each abnormal region only once, and determine a length of the abnormal region. Outputting the predictive pattern via an output interface in communication with the processor or storing the predictive pattern in memory, wherein the predictive pattern is a predictive estimate of an impending failure and assists in management of the machine.

First claim

What is claimed is:

1 . A system for determining a pattern in time series data representing an operation of a machine, comprising:
a sensor in communication with the machine; an output interface; a computer readable memory to store and provide a set of training data examples generated by the sensor in communication with the machine, wherein each training data example represents an operation of the machine for a period of time ending with a failure of the machine; a processor in communication with the computer readable memory, is configured to iteratively partition each training data example in the set of training data examples into a normal state region and an abnormal state region, determine a predictive pattern absent from the normal state regions and present in each abnormal state region only once, and determine a length of the abnormal state region, wherein each iteration the processor is configure to:
select a current time series length for the abnormal state region within each training data example beginning from an estimated moment in time when the machine entered an abnormal mode of operation, and ending at the moment of failure of the machine, wherein the current time series length is shortened starting from the start of time series to the end at the machine failure, by an increment of one-time step, per iteration, such that the current time series length is shorter than a previous current time series length for the abnormal state region selected for a previous iteration within the training data example;
partition each training data example in the set of training data examples into the normal state region and the abnormal state region having the current time series length;
identify a pattern in the set of training data examples, such that the pattern is different from any other patterns present in all normal state regions of the set of training data examples, and is similar to exactly one pattern in each abnormal state region of the set of training data examples; and
select the pattern as the predictive pattern, if the pattern is found; and
output the predictive pattern via an output interface in communication with the processor or storing the predictive pattern in the computer readable memory, wherein the predictive pattern is a predictive estimate of an impending failure and assists in management of the machine.

2 . The system of claim 1 , wherein the abnormal state region corresponds to the machine failing to operate normally ending with the failure of the machine, and the current time series length of the abnormal state region is an amount of discrete-time data within the abnormal state region of each training data example in the set of training data examples.

3 . The system of claim 1 , where the predictive pattern is different from a pattern in the normal region, if a Euclidean distance between the two patterns exceeds a pre-specified threshold.

4 . The system of claim 1 , where the predictive pattern is considered similar to a pattern in the normal region if a Euclidean distance between the two patterns is lower than a pre-specified threshold.

5 . The system of claim 1 , wherein the searching for the predictive pattern is performed by a fast shapelet discovery algorithm.

6 . The system of claim 1 , wherein the processor partitions each training data example in the set of training data examples into the normal state region based on data identifying a portion of the training data example that is generated by the sensor of machine while the machine was operating normally, and partitions each training data example in the set of training data examples into the abnormal state region based on data identifying a portion of the training data example that is generated by the sensor of machine while the machine was failing to operate normally and ending with the failure of the machine.

7 . The system of claim 1 , wherein a user interface in communication with the processor and the computer readable memory, acquires and stores the set of training data examples in the computer readable memory upon receiving an input from a surface of the user interface by a user.

8 . The system of claim 1 , wherein the periods of time for the set of training data examples are one of: the same period of time or some training data examples in the set of training data examples have periods of time that are different from other training data example periods of time in the set of training data examples.

9 . The system of claim 1 , further comprising:
receiving a test data example from a sensor in communication from a second machine and storing in the computer readable memory; determining, by the processor, based on at least one stored predictive pattern in the computer readable memory, whether one or more test data segment extracted from the test data example identifies a pattern of the second machine that corresponds to the at least one stored predictive pattern in the computer readable memory; selecting the pattern as a second predictive pattern, if the pattern is found; and storing the second predictive pattern in the computer readable memory or outputting the second predictive pattern via an output interface in communication with the processor, wherein the second predictive pattern assists in management of the second machine.

10 . The system of claim 9 , wherein the second machine is similar to the machine, and the sensor of the second machine measures a same parameter as a respective sensor of the sensor of the machine.

11 . The system of claim 10 , wherein each parameter relates to the operation of the machine including one or a combination of: fluid force data, fluid energy data, vibration data, temperature data, voltage data or current data.

12 . The system of claim 9 , wherein the test data example of the second machine has a same period of time as each training data example in the set of training data examples of the machine, or the test data example of the second machine has a same sampling rate as a regular sampling rate for each training data example in the set of training data example of the machine.

13 . A method for determining a pattern in time series data representing an operation of a machine, comprising:
accessing a set of training data examples generated by a sensor in communication with the machine stored in a computer readable memory, wherein each training data example represents an operation of the machine for a period of time ending with a failure of the machine; iteratively partitioning, by the computer, each training data example in the set of training data examples into a normal state region and an abnormal state region, determine a predictive pattern absent from the normal state regions and present in each abnormal state region only once, and determine a length of the abnormal state region, wherein each iteration includes:
selecting a current time series length for the abnormal state region within each training data example beginning from an estimated moment in time when the machine entered an abnormal mode of operation, and ending at the moment of failure of the machine, wherein the current time series length is shortened starting from the start of time series to the end at the machine failure, by an increment of one-time step, per iteration, such that the current time series length is shorter than a previous current time series length for the abnormal state region selected for a previous iteration within the training data example;
partitioning each training data example in the set of training data examples into the normal state region and the abnormal state region having the current time series length;
identifying

Assignees

• Mitsubishi Electric Res Laboratories Inc

Inventors

• Nikovski Daniel Nikolaev
• Zhu Yan
• Farahmand Amir-Massoud

Classifications

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G05B2219/24033

  Failure, fault detection and isolation · CPC title

• G05B23/0205 Primary

  by means of a monitoring system capable of detecting and responding to faults · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G05B23/0283

  Predictive maintenance, e.g. involving the monitoring of a system and, based on the monitoring results, taking decisions on the maintenance schedule of the monitored system; Estimating remaining useful life [RUL] (preventive maintenance, i.e. planning maintenance according to the available resources without monitoring the system G06Q10/06) · CPC title