Monitoring diagnosis device for electrical appliance

• US 5,305,235 A
• Filed: 07/09/1992
• Issued: 04/19/1994
• Est. Priority Date: 07/10/1991
• Status: Expired due to Fees

• Current Assignee: Korea Electric Power Corporation (Government of South Korea), Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation), Tokyo Electric Power Company
• Inventors: Izui, Yoshio, Akimoto, Yoshiakira, Tanaka, Hideo, Ogi, Hiromi
• Primary Examiner(s): Cosimano, Edward R.
• Application Number:
  US07/911,068
• Time in Patent Office: 649 Days
• Field of Search: 307/201, 307/464, 364/550, 364/551.01, 395/21, 395/23, 395/24, 395/50, 395/911, 395/912
• US Class Current: 702/185
• CPC Class Codes: G06F 11/22 Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: H02H 1/0023 sensing non electrical parameters, e.g. by optical, pneumatic, thermal or sonic sensors
• CPC Class Codes: H02H 1/0092 concerning the data processing means, e.g. expert systems, neural networks
• CPC Class Codes: H02H 7/22 for distribution gear, e.g. bus-bar systems; for switching devices detecting mechanical or electrical defects in gas-insulated switchgears H02B13/065
• CPC Class Codes: Y10S 706/911 Nonmedical diagnostics
• CPC Class Codes: Y10S 706/912 Manufacturing or machine, e.g. agricultural machinery, machine tool

Abstract

A monitoring diagnostic device for an electrical appliance such as gas insulated switchgear includes a sensor, such as an acceleration sensor, and a neural network including an input layer, an intermediate layer, and an output layer, each consisting of a plurality of neural elements. The input, intermediate and output layers are coupled to each other via a plurality of connection weights. The output of the sensor is first processed and then is supplied to the neural elements of the input layer. The connection weights are adjusted by means of learning data such that the output from the neural elements of the output layer of the neural network correctly identifies the causes of abnormality of the electrical appliance.

Claims

1. A monitoring diagnosis device for an electrical appliance, comprising:

sensor means for detecting a parameter of the electrical appliance and generating an output corresponding to the parameter indicative of a cause of abnormality of the electrical appliance said sensor means including a partial discharge sensor;

a neural network means including an input layer, an intermediate layer, and an output layer, the input layer, intermediate layer, and output layer each consisting of a plurality of neural elements each simulating a living neuron, wherein the neural elements of the input layer are coupled to the neural elements of the intermediate layer via respective connection weights, and the neural elements of the intermediate layer are coupled to the neural elements of the output layer via respective connection weights, and wherein said connection weights between the input layer and intermediate layer and between the intermediate layer and the output layer are adjusted on the basis of learning data consisting of causes of abnormality and instances of the output of said sensor means, such that a neural element of the output layer corresponding to a cause of abnormality has a high logic output in response to the output of said sensor means indicative of the existence of an abnormality while other neural elements of the output layer have a low logic output;

a preprocessor means for preprocessing output waveform samples of the output of said sensor means to obtain characteristic waveforms consisting of peaks each corresponding to an abrupt variation in the waveform samples;

an averaging means for averaging the characteristic waveforms to obtain an average characteristic waveform; and

normalizer means for normalizing a peak level of the averaged characteristic waveform to unity and dividing a time interval of the waveform into a plurality of subintervals, to obtain a characteristic waveform histogram;

wherein levels of said characteristic waveform histogram are input to corresponding neural elements of the input layer of said neural network means.

2. A monitoring diagnosis device for an electrical appliance as claimed in claim 1, wherein:

said neural network means comprises at least a first and a second neural network module;

said first module has a normal and an abnormal state output corresponding to a normal and an abnormal state of the electrical appliance, respectively, said first module including an input layer, an intermediate layer, and an output layer, and the input layer, intermediate layer, and output layer each consisting of a plurality of neural elements each simulating a living neuron, wherein the neural elements of the input layer are coupled to the neural elements of the intermediate layer via respective connection weights, and the neural elements of the intermediate layer are coupled to the neural elements of the output layer via respective connection weights, and the connection weights between the input layer and the intermediate layer and between the intermediate layer and the output layer are adjusted, on the basis of learning data consisting of information obtained from the output of said sensor means and a correspondence between the information obtained from the output of said sensor and normal and abnormal states of the electrical appliance, such that said normal state output has a high logic level in response to a normal state of the electrical appliance and said abnormal state output has a high logic level in response to an abnormal state of said electrical appliance; and

said second module includes an input layer, an intermediate layer, and an output layer, the input layer, intermediate layer, and output layer each consisting of a plurality of neural elements each simulating a living neuron, wherein the neural elements of the input layer are coupled to the neural elements of the intermediate layer via respective connection weights, and the neural elements of the intermediate layer are coupled to the neural elements of the output layer via respective connection weights, and said connection weights between the input layer and the intermediate layer and between the intermediate layer and the output layer are adjusted on the basis of learning data including causes of abnormality and instances of the output of said sensor means, such that a neural element corresponding to a cause of abnormality has a high logic level in response to the output of said sensor means, wherein said second module has an input coupled to said abnormal state output of said first module and is activated in response to an assertion of abnormal state output of said first module.

3. A monitoring diagnosis device for an electrical appliance as claimed in claim 1, wherein:

said neural network means comprises at least a first, a second, a third and a fourth neural network module, each including an input layer, an intermediate layer, and an output layer, and the input layer, intermediate layer, and output layer each consisting of a plurality of neural elements each simulating a living neuron, wherein the neural elements of the input layer are coupled to the neural elements of the intermediate layer via respective connection weights, and said neural elements of said intermediate layer are coupled to said neural elements of said output layer via respective connection weights;

said first module has a normal and an abnormal state output corresponding to a normal and an abnormal state of the electrical appliance, respectively, and the connection weights between the input layer and the intermediate layer and between the intermediate layer and the output layer are adjusted, on the basis of learning data consisting of information obtained from the output of said sensor means and a correspondence between said information and normal and abnormal states of said electrical appliance, such that said normal state output has a high logic level in response to an output of said sensor means indicative of a normal state of the electrical appliance and said abnormal state output has a high logic level in response to an output of said sensor means indicative of an abnormal state of the electrical appliance;

said second module has an input coupled to the abnormal state output of said first module and said second module being activated in response to a high logic abnormal state output of said first module, and said second module having at least a first and a second output corresponding to a first and a second group of causes of abnormality of the electrical appliance, the connection weights between the input layer and the intermediate layer and between said the intermediate layer and the output layer of said second module are adjusted on the basis of learning data consisting of a plurality of groups of causes of abnormality and instances of the output of said sensor means, such that an output corresponding to a group of causes of abnormality has a high logic level in response to the output of said sensor means;

said third module has an input coupled to the first output of said second module, said third module being activated in response to a high logic level of said first output of said second module, wherein the connection weights between the input layer and the intermediate layer and between the intermediate layer and the output layer of said third module are adjusted on the basis of learning data consisting of a first group of the plurality of causes of abnormality and instances of the output of said sensor means, such that a neural element corresponding to a cause of abnormality belonging to the first group of causes of abnormality has a high logic level in response to the output of said sensor means; and

said fourth module has an input coupled to the second output of said second module, said fourth module being activated in response to a high logic level of the second output of said second module, wherein the connection weights between the input layer and the intermediate layer and between the intermediate layer and the output layer of said fourth module are adjusted on the basis of learning data consisting of a second group of the plurality of causes of abnormality and instances of the output of said sensor means, such that a neural element corresponding to a cause of abnormality belonging to the second group of causes of abnormality has a high logic level output in response to the output of said sensor means.

4. A monitoring diagnosis device for an electrical appliance, comprising:

an acceleration sensor which detects mechanical vibrations and generates an output waveform;

a Fast Fourier Transform means for generating spectrum curves of the output waveform by performing a Fast Fourier Transform on the waveform;

means for averaging the spectrum curves to obtain an averaged spectrum curve;

first means for generating a histogram of the averaged spectrum curve;

a partial discharge sensor for detecting variations in potential and for generating an output waveforms;

a preprocessor means for preprocessing the output waveforms of said partial discharge sensor to obtain characteristic waveforms consisting of a plurality of peaks;

an averaging means for averaging the characteristic waveforms to obtain an averaged characteristic waveform; and

second means for generating a histogram of the averaged characteristic waveform; and

a neural network having first and second inputs connected to said first and second means for generating a histogram, said neural network including an input layer, an intermediate layer, and an output layer, the input layer, intermediate layer, and output layer each consisting of a plurality of neural elements each simulating a living neuron, wherein the neural elements of the input layer are coupled to the neural elements of the intermediate layer via respective connection weights, and the neural elements of the intermediate layer are coupled to the neural elements of the output layer via respective connection weights, and wherein said connection weights between the input layer and intermediate layer and between the intermediate layer and the output layer are adjusted on the basis of learning data consisting of causes of abnormality and instances of the output of said acceleration sensor and said partial discharge sensor, such that a neural element of the output layer corresponding to a cause of abnormality has a high logic output in response to the output waveform of said acceleration sensor and in response to the output waveform of said partial discharge sensor indicative of the existence of an abnormality while other neural elements of the output layer have a low logic output.

5. A monitoring diagnosis device for an electrical appliance as claimed in claim 4, wherein:

said neural network comprises at least a first and a second neural network module;

said first module having a normal and an abnormal state output corresponding to a normal and an abnormal state of the electrical appliance, respectively, said first module including an input layer, an intermediate layer, and an output layer, and the input layer, intermediate layer, and output layer each consisting of a plurality of neural elements; and

said second module including an input layer, an intermediate layer, and an output layer, the input layer, intermediate layer, and output layer each consisting of a plurality of neural elements, said second module having an input coupled to said abnormal state output of said first module.