RU2580183C1 - Method for diagnosing electric switching device - Google Patents

Abstract

FIELD: measuring equipment.

SUBSTANCE: invention can be used for control of serviceability of electric switching devices, mainly of high-voltage automatic circuit breakers. Disclosed is a diagnostic technique for electric switching device based on measuring function of the movable contact during connection and/or disconnection of the diagnosed and deciding on the state of the apparatus. Movable contact is taken dependence of acceleration time are selected on the dependence of high-frequency component to generate several frequency bands change of high-frequency component. From the obtained frequency bands using neural networks-“analysers” and informative parameter K_q is separated as diagnostic signs of the most informative frequency bands. Then the obtained values of the selected frequency ranges of the measured acceleration function of time of the movable contact, the most informative diagnostic features, is fed to the input of the neural network-“expert”, which decides on the character of failure of switching device.

EFFECT: technical result consists in possibility of automatic diagnostics of high-voltage breakers with indication of cause of failure.

1 cl, 5 dwg

Description

The invention relates to the field of measurement and control and can be used to control the serviceability of electric circuit breakers.

A known method for diagnosing an electrical switching device [Patent RU No. 2117309, IPC G01R 31/333, publication date: 08/10/1998], in which the capacitance between the contacts of the device in the off state and in the measurement coordinate or in two coordinates adjacent to it is determined, form the first informative parameter characterizing the change in capacitance in the vicinity of the coordinate of measurement according to the first embodiment, in the form of the first derivative of the capacitance with respect to the coordinate, and in the second embodiment, in the form of the difference of the mentioned nearby coordinates, measure e the capacitance between the contacts of this apparatus and is stored in the form of a dependence of the capacitance on time, the moment of time for scaling the capacitive dependencies of the reference and diagnosed apparatuses is selected, the capacitance between the contacts of the diagnosed apparatus at the time instant corresponding to the passage of the point with the coordinate of the measurement is determined. According to the dependence of the capacitance on time, a second informative parameter is formed, characterizing its change in the vicinity of the measurement coordinate according to the first embodiment — in the form of the first derivative of this capacitance with respect to time, and according to the second variant — in the form of the interval of time the contact moves between the mentioned nearby coordinates, and the moving speed contact judged by the ratio of informative parameters.

The disadvantage of this method is the inability to identify the causes of malfunction of the switch.

There is also a method for diagnosing electrical switching devices using a vibrograph [Collection of teaching aids for monitoring the status of electrical equipment. Section 4. Methods for monitoring the status of switching devices. - M.: ORGRES, 1997, p. 243-245], based on the registration of the speed and movement of the moving parts of the high voltage switch, in which the waveform carrier is mounted on a rod attached to the movable part of the high voltage switch, and the recording element of the waveform is attached to the fixed housing of the high voltage switch.

The disadvantage of this method for diagnosing electrical switching devices is the need to manually process vibrograms and the difficulties of its automation, the inability to automatically log diagnostic results and identify the causes of breaker breakdowns.

Also known is a method for early detection of defects in the mechanisms of high-voltage circuit breakers [Chernyshev N.A., Rakevich A.L. The apparatus and method for early detection of defects in the mechanisms of high-voltage circuit breakers - Electrical stations, 2004, No. 11, p. 61-65], which consists in measuring the moving function of the movable contact of the high voltage circuit breaker in the form of a speed characteristic using a displacement sensor, dividing it into temporary sections, where the influence of various parts of the high voltage circuit breaker is most manifested, then analyzing the obtained graphs by comparing the measurement results with permissible values and deciding on the state of the high voltage circuit breaker.

This method is the closest to the claimed technical essence and is taken as a prototype.

The disadvantages of the prototype method include the lack of automatic decision-making on the state of the high-voltage switch indicating the cause of the malfunction, as a result, the low reliability of the diagnostic process due to subjective decision-making by the operator.

An object of the invention is to provide a method for diagnosing an electrical switching apparatus, mainly a high-voltage circuit breaker (BB), providing the ability to automatically diagnose high-voltage circuit breakers with an indication of the cause of the malfunction.

The problem is solved using the neural network of the "expert", trained in the library of serviceable and faulty characteristics of explosives. However, before using the “expert” neural network, the analyzed characteristics of explosives must be specially converted in order to isolate certain informative parameters from them for automatic reliable assessment of the state of explosives. Studies have shown that the use and training of an “expert” neural network using the speed characteristics used in the prototype method makes it possible to achieve a diagnostic reliability of not more than 80%. Thus, to significantly increase the reliability of the diagnosis of explosives, it is necessary to use a different sequence of operations than in the prototype method in the process of converting the controlled parameters of explosive motion.

An increase in the level of reliability in the proposed method is possible due to the formation of a special informative parameter K and the allocation and use for diagnosis on its basis of the most informative frequency ranges of the acceleration function of the movable contact of explosives.

In the claimed technical solution, the dependence of its acceleration on time is taken as the mentioned function of moving the movable contact, the high-frequency component is distinguished on this dependence, several frequency ranges of the high-frequency component are formed, a neural network “analyzer” is formed for each of the formed frequency ranges, an informative parameter is determined K _q for each generated frequency range according to the formula:

where q = 1 ... p; p is the number of generated frequency ranges; j = 1 ... m; m is the number of discrete signal values in the qth frequency range; d _ql , d _qt - reference values of the output signal of the neural network “analyzer” for the q-th frequency range during training and testing, respectively; φ _ql , φ _qt - activation functions for the q-th frequency range during training and testing, respectively; x _qj is the jth value of the signal level inside the qth frequency range; w _qj is the weight coefficient for the j-th signal level value inside the q-th frequency range, compare the level of each obtained value of the informative parameter K _q with its predetermined threshold value, select the most informative frequency ranges for which the informative parameter K _q did not exceed the threshold value, the values of the allocated frequency ranges of the measured acceleration function versus the time of the moving contact, the most informative as diagnostic signs, are fed in one neural network is an “expert”, and the nature of the malfunction of the switching apparatus is judged by the value of the signals at the outputs of the neural network “expert”.

Thus, at the stage of comparing the acceptable values of the signal with the measured values, the high-frequency components of the acceleration function are used with the frequency ranges that are most informative as diagnostic signs and allocated using the informative parameter K, as a result, the decision on the technical condition of the explosive and the cause of its malfunction is made not by the operator, but neural network is an “expert."

The proposed solution is illustrated in FIG. 1-5. In FIG. 1 is a structural diagram of a variant of a diagnostic system that implements the proposed method, which includes the following elements: microcontroller 1, interface module with power contacts 2, interface module with sensors 3, analog acceleration sensor 4, display 5, flash drive 6.

The microcontroller 1 controls the entire device as a whole, receiving and processing measurement information. The module for interfacing with power contacts 2 carries out galvanic isolation with the power circuits of the circuit breaker being diagnosed and supplying logic signals corresponding to the current state of contacts of each phase (closed / open) to the corresponding input / output lines of the microcontroller and the presence of a start pulse of the circuit breaker drive, along the edge of which the process begins acceleration measurements. The sensor interface module 3 is used to connect an analog acceleration sensor. The inputs of the microcontroller 1 are connected to the outputs of the interface module with power contacts 2 and the interface module with sensors 3, as well as display 5 and flash drive 6 for displaying and saving diagnostic results.

In FIG. 2 is a structural diagram of an “expert” neural network 7, which can be various types of neural networks, such as a perceptron. In FIG. Figure 3 shows three functions of the explosive contact speed as a function of its course: one operational (curve 8) and two faulty explosives (curve 9 and curve 10). In FIG. Figure 4 shows three graphs of the change in the level of the acceleration signal in different frequency ranges: one healthy (photo 11) and two faulty explosives (photo 12 and 13), each of which is a high-frequency component of the acceleration signal a (t) divided into 40 frequency ranges. In FIG. Figure 5 shows the dependences of accuracy (a parameter that complements the diagnostic error up to 100%) of a neural network on the parameters of the learning speed λ and the number of epochs of learning the neural network.

Consider the method of diagnosing an electrical switching device using the example of a system that implements the proposed method (Fig. 1). At the first stage, using the analog acceleration sensors 4 and the interface module with sensors 3, the acceleration function is measured in time of the moving explosive contacts when they are closed / opened (on / off). Determining the start and end of the measurement provides a module for interfacing with power contacts 2. From the obtained acceleration function, the microcontroller 1 selects its high-frequency component. In it, in accordance with the proposed method, the signal of the selected high-frequency component is divided into several frequency ranges p. Then p neural networks are formed - “analyzers” of the perceptron type (one for each frequency range). Using neural networks, "analyzers" form an informative parameter K for each frequency range according to the formula:

$$K_q=\left|\left(d_{ql}-{\varphi }_{ql}\left({\displaystyle \sum _{j=0}^m{w}_{qj}\cdot x_{qj}}\right)\right)-\left(d_{qt}-{\varphi }_{qt}\left({\displaystyle \sum _{j=0}^m{w}_{qj}\cdot x_{qj}}\right)\right)\right|,\left(\mathrm{one}\right)$$

where q = 1 ... p, p is the number of generated frequency ranges; j = 1 ... m, m is the number of discrete signal values in the qth frequency range; d _ql , d _qt are the reference values of the output signal of the neural network “analyzer” for the q-th frequency range during training and testing, respectively; φ _ql , φ _qt - activation functions for the q-th frequency range during training and testing, respectively; x _qj is the jth value of the signal level inside the qth frequency range; w _qj is the weight coefficient for the j-th signal level value inside the q-th frequency range.

The informative parameters K formed using neural networks-“analyzers” are further compared with a set threshold value. If the informative parameter K _{q is} higher than the specified threshold value, then the corresponding qth frequency range is removed from the spectrum of the high-frequency acceleration function. The higher the threshold level for the informative parameter K, the more informative ranges will remain in the high-frequency component of the acceleration function. The last stage in diagnosing the state of explosives by the function of accelerating their moving contacts is the launch of an “expert” neural network 7, which decides on the state of explosives. The input of the “expert” neural network 7 (Fig. 2) receives the obtained values of y ₁ -y _{n of the} selected frequency ranges, the most informative as diagnostic features. As a result of the work, the neural network “expert” 7, trained on the corresponding reference values of the same given informative frequency ranges and decides on the presence of corresponding defects in the switching apparatus, makes a decision on its technical condition. The microcontroller 1 displays the results of the procedure on display 5 and saves it to the flash drive 6.

The advantages of using this method are illustrated in FIG. 3-5.

The graphs of FIG. 3-4 show that dividing the high-frequency component of the signal into frequency ranges allows you to set clear boundary informative parameters that distinguish between healthy and faulty circuit breakers, since there are big differences between the performance graphs of a healthy and faulty high-voltage circuit breaker.

To illustrate the advantages of the proposed allocation of the most informative frequency ranges obtained using the informative parameter K, an “expert” neural network was constructed, which was trained on the frequency ranges of the high-frequency component of the acceleration signal using the operation of extracting the most informative ranges and without it.

Graphs 14 and 15 show the dependencies of accuracy, respectively, during training and testing before applying the operation to select the information range described above, and the corresponding graphs 16 and 17 after applying this operation.

As can be seen from graphs 14, 15, 16, 17 in FIG. 5, with various combinations of parameters of the “expert” neural network (learning speed, number of epochs), its adequacy increased, i.e. the accuracy graphs of the “expert” neural network during training and testing ceased to differ and became almost identical. At the same time, the accuracy of the “expert” neural network was achieved, with testing up to 90%, which in this case is a parameter opposite to the diagnostic error. The data size for training has also decreased from 40 frequency ranges of the wavelet spectrum to 11 of the most informative ones, which significantly speeds up the learning process of an “expert” neural network.

The most appropriate application of the proposed technical solution for diagnostics both in the production process and during operation at electrical substations of high-voltage circuit breakers of various types and modifications with an operating voltage of 10 kV to 220 kV.

Claims (1)

A method for diagnosing an electrical switching apparatus, based on the fact that, at least in one phase, the function of moving a movable contact is measured when the diagnosed apparatus is turned on and / or off, the measurement results are compared with acceptable values and a decision is made on the state of the apparatus, characterized in what is taken as the mentioned function of moving the movable contact is the dependence of its acceleration on time, a high-frequency component is isolated on this dependence, several parts are formed tnyh band high-frequency component changes, neural Network-formed "analyzer" formed for each frequency band determined by K _q informative parameter for each frequency band generated by the formula:

where q = 1 ... p;
p is the number of generated frequency ranges;
j = 1 ... m;
m is the number of discrete signal values in the qth frequency range; d _ql , d _qt - reference values of the output signal of the neural network “analyzer” for the q-th frequency range during training and testing, respectively;
φ _ql , φ _qt - activation functions for the q-th frequency range during training and testing, respectively;
x _qj is the jth value of the signal level inside the qth frequency range;
w _qj is the weight coefficient for the j-th signal level inside the q-th frequency range;
the level of each obtained value of the informative parameter K _{q is} compared with its predetermined threshold value, the most informative frequency ranges for which the informative parameter K _q does not exceed the threshold value are selected as diagnostic signs, the values of the selected frequency ranges of the measured acceleration function from the time of the moving contact are fed, the most informative as diagnostic signs, to the input of the neural network - “expert”, and the nature of the malfunction of the switching apparatus is judged by the value of the signals at the outputs of the neural network - "expert".

Images