Method and testing device

Patent metadata

Field	Value
Publication number	US-11415618-B2
Application number	US-201917429036-A
Country	US
Kind code	B2
Filing date	Feb 6, 2019
Priority date	Feb 6, 2019
Publication date	Aug 16, 2022
Grant date	Aug 16, 2022

Abstract

A method for testing a network is disclosed, the network having a number of network sections, in particular in a cable harness having a number of such networks, having the following steps of: recording training measured values for a number of reference networks, wherein the reference networks correspond to the network to be tested, preprocessing the recorded training measured values in order to eliminate data errors in the training measured values, training a first classification system using the training measured values, wherein the first classification system is based on at least one algorithm from the field of machine learning and is designed to classify a network either as fault-free or faulty, training a second classification system using the training measured values, wherein the second classification system is based on at least one algorithm from the field of machine learning and is designed to classify a faulty network section of a network, recording test measured values for the network to be tested, preprocessing the recorded test measured values in order to eliminate data errors in the training measured values, classifying the network to be tested as fault-free or faulty on the basis of the recorded test measured values using the trained first classification system, and classifying the faulty network section of the network to be tested using the trained second classification system if the network (to be tested was classified as faulty by the trained first classification system. The present invention also discloses a corresponding testing device.

First claim

The invention claimed is:

1. A method for testing a network comprising a number of network sections or a line set comprising a number of such networks, the method comprising the steps of:
recording training measured values for a number of reference networks, the reference networks corresponding to the network to be tested,
preprocessing the recorded training measured values to eliminate data errors in the training measured values,
training a first classification system with the training measured values, the first classification system being based on at least one algorithm from the field of machine learning and designed to form a network either to be classified as faulty-free or faulty,
training a second classification system with the training measured values, the second classification system being based on at least one algorithm from a field of machine learning and being designed to identify a faulty network section of a network,
recording test measured values for the network to be tested,
preprocessing the recorded test measured values in order to eliminate data errors in the training measured values,
classifying the network to be tested as faulty-free or faulty on the basis of the recorded test measured values with the trained first classification system, and
classifying the faulty network section of the network to be tested with the trained second classification system if the network to be tested has been classified as faulty by the trained first classification system.

2. The method according to claim 1 ;
wherein the recording of training measured values comprises the recording of measured values on faulty-free reference networks, and
wherein the recording of training measured values comprises the recording of measured values on reference networks in which errors were generated at predetermined network sections.

3. The method according to claim 2 , wherein, in order to generate the errors in the corresponding network sections, the dielectric constant is changed by a corresponding line of the respective network.

4. The method according to claim 1 , wherein:
the preprocessing of the recorded training measured values comprises recognizing and eliminating outliers in the recorded training measured values, and/or
the preprocessing of the recorded test measured values comprises recognizing and eliminating outliers in the recorded test measured values.

5. The method according to claim 4 , wherein the detection of outliers comprises recognizing and treating local outliers or global outliers.

6. The method according to claim 4 , wherein the recognition of outliers is carried out by means of the 2-sigma rule or by means of an outlier test according to Grubbs and Hampel or by means of a local outlier factor algorithm.

7. The method according to claim 1 , wherein:
the preprocessing of the recorded training measured values comprises detecting and eliminating offset errors in the recorded training measured values, and/or
preprocessing of the recorded test measured values comprises detecting and eliminating offset errors in the recorded test measured values.

8. Method according to claim 7 , wherein:
the detection and elimination of offset errors comprises converting the recorded training measured values into differential measured values, and/or
wherein the detecting and eliminating of offset errors comprises converting the recorded test measured values into differential measured values.

9. The method according to claim 1 , wherein:
the preprocessing of the recorded training measured values comprises aggregating a plurality, 2, 4, 8, 16, 32 or 64 individual measurement curves and/or
the preprocessing of the recorded test measured values comprises aggregating in each case a plurality, 2, 4, 8, 16, 32 or 64 individual measurement curves.

10. The method according to claim 1 , wherein:
the first classification system is based on at least one of a decision tree algorithm, an algorithm according to the ensemble method, an AdaBoost algorithm, an algorithm for logistic regression, a Naive Bayes classifier algorithm, a K-nearest-neighbor classifier algorithm or a support vector machine algorithm based, and/or
the second classification system based on at least one of a decision tree algorithm, an algorithm according to the ensemble method, an AdaBoost algorithm, an algorithm for logistic regression, a Naive Bayes classifier algorithm, a K-nearest-neighbor classifier algorithm or a support vector machine algorithm based.

11. The method according to claim 10 , wherein:
for the first classification system a parameter optimization is performed, and/or
a parameter optimization is carried out for the second classification system.

12. The method according to claim 1 , wherein:
during training the first classification system, at least one of a predetermined proportion or between 70% and 95% of the preprocessed training measured values is used for training the first classification system, and the remaining proportion of the preprocessed training measured values is used for verification of the training, and/or
wherein, during training of the second classification system, at least one of a predetermined proportion or between 70% and 95% of the preprocessed training measured values is used for training the first classification system and the remaining proportion of the preprocessed training measured values is used for verification of the training.

13. A testing apparatus for testing a network having a number of network sections or in a line set having a number of such networks, comprising:
a first data acquisition device configured to record training measured values for a number of reference networks, wherein the reference networks correspond to the network to be tested,
a computing device configured to preprocess the recorded training measured values in order to eliminate data errors in the training measured values,
a first classification system based on at least one algorithm from the field of machine learning and designed to classify a network as either faulty-free or faulty,
wherein the computing device is further configured to train the first classification system with the training measured values,
a second classification system based on at least one algorithm from the field of machine learning and designed to classify a faulty network section of a network,
wherein the computing device is further configured to train the second classification system with the training measured values,
second data acquisition device which is designed to record test measured values for the network to be tested, and
a test control device configured to preprocess the recorded test measured values in order to eliminate data errors in the training measured values,
wherein the test control device is further configured to classify the network ti be tested as faulty-free or faulty on the basis of the recorded test measured values with the trained first classification system, and
wherein the test control device is further configured to classify the faulty network section of the network to be tested with the trained second classification system if the network to be tested has been classified as faulty by the trained first classification system.

14. The test apparatus according to claim 13 , wherein
the computing device is configured to record measured values on faulty-free reference networks when recording training measured values, and/or
the computing device being designed to record training measured values, on reference networks when recording training measured values, in which faults have been generated at predetermined network sections, in ord

Assignees

• Draexlmaier Lisa Gmbh

Inventors

• Franke Michael
• Kappl Andreas

Classifications

• G01R31/086 Primary

  in power transmission or distribution networks, i.e. with interconnected conductors · CPC title

• G01R31/007 Primary

  using microprocessors or computers · CPC title

• B60R16/0207

  Wire harnesses · CPC title

• G01R31/52

  Testing for short-circuits, leakage current or ground faults · CPC title

• G01R31/088

  Aspects of digital computing · CPC title