System and signal processing method for failure mode identification

Patent metadata

Field	Value
Publication number	US-2022012116-A1
Application number	US-202016925585-A
Country	US
Kind code	A1
Filing date	Jul 10, 2020
Priority date	Jul 10, 2020
Publication date	Jan 13, 2022
Grant date	—

Abstract

Systems and methods described herein are directed to minimizing the resource requirements for edge and network while keeping the accuracy of machine learning classifier by utilizing simulated test data. Once sufficient measured test data is collected by the server, the server instructs the edge computer to reduce the transmission of data received from the corresponding sensors.

First claim

What is claimed is:

1 . A system, comprising:
an edge computer configured to process and classify sensor data from a sensor; and a server, comprising:
a processor, configured to:
accept measured normal mode features from the edge computer to train a classifier;
produce simulated failure modes features from a simulator to train the classifier;
train the classifier with the measured normal features and simulated failure mode features; and
instruct the edge computer to reduce data transmission to the server in response to a determination that a classification accuracy of the trained classifier is more than a threshold.

2 . The system of claim 1 , wherein the processor is configured to instruct the edge computer to halt the data transmission to the server.

3 . The system of claim 1 , wherein the edge computer is configured to be triggered to activate transmission of data from the edge computer to the server.

4 . The system of claim 1 , wherein the sensor data comprises wideband sensor signal, and wherein the edge computer is configured to reduce the data transmission to the server by limiting bandwidth of the wideband sensor signal according to an expected bandwidth of a failure mode determined by the server

5 . The system of claim 1 , wherein the sensor data comprises vibration data and wherein the edge computer is configured to instruct the server to reinitiate training of the classifier for when the vibration data falls below a threshold.

6 . The system of claim 1 , wherein the edge computer is configured to process the sensor data from the sensor through generating a 2-D matrix of features from concatenation of statistics of spectrum data in the sensor data, the 2-D matrix of features configured to be incorporated by a 2-D input neural network implemented by the classifier of the server.

7 . A method for a system comprising an edge computer configured to process and classify sensor data from a sensor, and a server, the method comprising:
accepting, at the server, measured normal mode features from the edge computer to train a classifier; producing simulated failure modes features from a simulator to train the classifier, training the classifier with the measured normal features and simulated failure mode features; and instructing the edge computer to reduce data transmission to the server in response to a determination that a classification accuracy of the trained classifier is more than a threshold.

8 . The method of claim 7 , further comprising instructing the edge computer to halt the data transmission to the server.

9 . The method of claim 7 , wherein the edge computer is configured to be triggered to activate transmission of the data from the edge computer to the server.

10 . The method of claim 7 , wherein the sensor data comprises wideband sensor signal, and wherein the edge computer reduces the data transmission to the server by limiting bandwidth of the wideband sensor signal according to an expected bandwidth of a failure mode determined by the server.

11 . The method of claim 7 , wherein the sensor data comprises vibration data and wherein the method further comprises instructing the server to reinitiate training of the classifier for when the vibration data falls below a threshold.

12 . The method of claim 7 , wherein the edge computer is configured to process the sensor data from the sensor through generating a 2-D matrix of features from concatenation of statistics of spectrum data in the sensor data, the 2-D matrix of features configured to be incorporated by a 2-D input neural network implemented by the classifier of the server.

13 . An edge computer configured to process and classify sensor data from a sensor for transmission to a server, the edge computer comprising:'
a processor, configured to:
provide measured normal mode features to the server to train a classifier; and
for receipt of instructions from the server to reduce the data transmission to the server, reduce the data transmission to the server.

14 . The edge computer of claim 13 , wherein the processor is configured to reduce the data transmission to the server through halting the data transmission to the server.

15 . The edge computer of claim 13 , wherein the edge computer is configured to be triggered to activate transmission of data from the edge computer to the server.

16 . The edge computer of claim 13 , wherein the sensor data comprises wideband sensor signal, and wherein the processor is configured to reduce the data transmission to the server by limiting bandwidth of the wideband sensor signal according to an expected bandwidth of a failure mode determined by the server.

17 . The edge computer of claim 13 , wherein the sensor data comprises vibration data and wherein the processor is configured to instruct the server to reinitiate training of the classifier for when the vibration data falls below a threshold.

18 . The edge computer of claim 13 , wherein the processor is configured to process the sensor data from the sensor through generating a 2-D matrix of features from concatenation of statistics of spectrum data in the sensor data, the 2-D matrix of features configured to be incorporated by a 2-D input neural network implemented by the classifier of the server.

Assignees

• Hitachi Ltd

Inventors

• Maeda Daisuke
• Gaur Sudhanshu

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0895

  Weakly supervised learning, e.g. semi-supervised or self-supervised learning · CPC title

• G06N3/096

  Transfer learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• H04W24/02 Primary

  Arrangements for optimising operational condition · CPC title