JP2006250818A - Diagnosis system for electrical installation - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a diagnosis system for electrical installation, capable of diagnosing a winding section from the shape of waveform pattern to which the hidden Markov model has been applied to the waveform pattern of characteristic quantity obtained by impressing a specific impulse in the winding section of electrical installation, and determining. <P>SOLUTION: The specific impulse is impressed from an impulse-generating circuit 3 to the winding section 2 of the electrical equipment, and the characteristic quantity detected with a characteristic quantity detector 4 is input in a learning section 6 by way of an A/D conversion circuit 5. The learning section 6 makes the waveform patterns of each characteristic quantity correspond to each state of normal, failure and degradation of the winding section 2 into parameters based on the hidden Markov model and stores. When diagnosing the winding section 2, the impulse with the same characteristic as the specific impulse is impressed from the impulse generation circuit 3 to the winding section 2. A diagnosis section 7 calculates the formation probability of the waveform pattern of the detected characteristics quantity, based on the parameters stored in the learning section 6, determines the parameter whose formation probability becomes maximum, decides the state of the winding section 2 and indicates the results on a display 8. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an electrical equipment diagnosis system for diagnosing the state of a winding part constituting a rotating machine (an electric motor, a generator, etc.) or a transformer provided in a factory or a building using a statistical pattern recognition method. .

Conventionally, when diagnosing a winding part of a rotating machine or a transformer in a factory or a building, an impulse test is performed in which an impulse is applied to the winding part to be diagnosed. In this case, the voltage waveform detected from the winding portion is set by setting the winding portion to a normal state or an assumed abnormal state (failure state) and applying a predetermined impulse to the winding portion in each state. The area of the voltage waveform in each state is stored in advance in the storage unit. Next, at the time of diagnosis of the winding part, the area of the voltage waveform obtained when an impulse having the same characteristics as the predetermined impulse is applied to the winding part, and the winding stored in advance in the storage unit described above. The area of the voltage waveform in each state of the line part is compared. Thereby, it is diagnosed whether the winding part is in a normal state or an abnormal state (see Non-Patent Documents 1 and 2).
FIG. 6 shows the waveform of the observed voltage from the application period (period from point A to point B) to the decay period (after point B) when an impulse is applied to the winding part to be diagnosed. In the conventional winding section diagnosis, the area of the voltage waveform during the decay period when the waveform is relatively stable is compared to diagnose whether the winding section is in a normal state or an abnormal state. If it is in a state, it is determined what kind of abnormality it is.

In the case of the conventional winding part diagnosis means, as described above, by comparing the areas of the voltage waveforms, it is diagnosed whether the winding part is in a normal state or an abnormal state. It is judged whether it is abnormal. However, even when the observed voltage waveforms are different corresponding to the respective abnormal states of the winding portions, the areas of the voltage waveforms may be the same. In such a case, it is impossible to accurately determine what abnormality (failure) the winding portion has.
Proceedings of the 2003 Annual Meeting of the Institute of Electrical Engineers of Japan "Judgment Method in Line Testing Machines" Electronic Control International Co., Ltd. “Impulse Winding Tester DXW-01, 05”

  Therefore, in the present invention, the state of the winding portion is determined from the shape of the waveform pattern in which the hidden Markov model is applied to the waveform pattern of the feature amount obtained by applying a predetermined impulse to the winding portion constituting the electrical equipment. It is a problem to be solved to provide an electrical equipment diagnosis system that can accurately diagnose and determine. The impulse applied to the winding part to be diagnosed is connected to the impulse output from the impulse generating circuit or the winding part as described in the section of the best mode for carrying out the invention described later. The counter electromotive force by the counter electromotive force induction circuit is also included.

The above problems can be solved by the electrical equipment diagnosis system described in the appended claims.
According to the electrical equipment diagnosis system according to claim 1, the learning means is obtained when a predetermined impulse is applied to the winding part in each predetermined state of the winding part constituting the electrical equipment. Each parameter obtained by parameterizing the waveform pattern of the feature quantity based on the hidden Markov model is stored. Further, the diagnosis means stores the generation probability of the waveform pattern of the feature amount detected in the state in which the impulse having the same characteristic as the predetermined impulse is applied to the winding when the winding is diagnosed. The calculation is performed based on the respective parameters, and the state of the winding portion is diagnosed and determined based on the parameter having the maximum generation probability. Thereby, the state of the winding portion is accurately diagnosed and determined.

  According to the electrical equipment diagnosis system of claim 2, the state of the winding part is roughly determined from the frequency spectrum waveform of the feature quantity or the time-series waveform of the frequency spectrum by the frequency analysis means. Can do. In addition to this, the learning means and the diagnostic means can diagnose and determine the winding portion, so that the state of the winding portion can be more accurately diagnosed and determined.

  Furthermore, according to the electrical equipment diagnostic system of the third aspect, since the parameters stored in advance in the learning means can be updated or exchanged, the versatility of the electrical equipment diagnostic system can be enhanced.

Here, the concept of the hidden Markov model will be described with reference to the schematic diagram shown in FIG.
In FIG. 1, a11, a12, a22, a23, a33, a34 indicate state transition probabilities, and aij is the probability of transition from state Si to state Sj. Further, bi (x) indicates the probability of outputting the feature quantity x in the state Si. S1 indicates an initial state, and S4 indicates a final state. As described above, the hidden Markov model has the state transition probability, the output probability, and the initial state probability πi (probability that the initial state is Si) as parameters, and each parameter is a normal state of the winding part constituting the electrical equipment. Or stored for each failure, abnormality, or deterioration state to be recognized, and corresponds to the function of the learning means described above.

  According to the present invention, whether the winding part is normal from the shape of the waveform pattern to which the hidden Markov model is applied to the waveform pattern of the characteristic amount obtained by applying the impulse to the winding part constituting the electrical equipment, It is possible to accurately diagnose and determine whether it is a failure or abnormality or a deteriorated state.

Next, an embodiment of the present invention will be described.
FIG. 2 is a block diagram showing the overall configuration of the electrical equipment diagnosis system 1. The electrical equipment diagnosis system 1 includes a winding unit 2 that constitutes a rotary machine such as an electric motor and a generator, a transformer, and the like. Diagnose and judge.
As shown in FIG. 2, the electrical equipment diagnosis system 1 is provided with an impulse generation circuit 3 for generating a predetermined impulse applied to the winding part 2 to be diagnosed. In addition, a feature amount detection unit 4 is provided for detecting a feature amount obtained when a predetermined impulse is applied to the winding portion 2, for example, a voltage, a current, or the like. Since the feature quantity detected by the feature quantity detection unit 4 is generally an analog value, an A / D conversion circuit 5 for converting the analog signal into a digital signal is provided. The A / D conversion circuit The digital signal output from 5 is input to the learning unit 6 described below in a predetermined state of the winding unit 2, for example, a normal state, a failure / abnormality, a deterioration state, and the like. At the time of diagnosis of the line part 2, it is inputted to the diagnosis part 7.

  The learning unit 6 is characterized in that the respective predetermined states of the winding unit 2 are set and a predetermined impulse is applied to the winding unit 2 in a state where driving power is not supplied to the electrical equipment. The signal after the various feature quantities detected by the quantity detection unit 4 are converted into digital signals by the A / D conversion circuit 5 is input, and the waveform pattern of each feature quantity is parameterized based on the hidden Markov model. Each parameter is stored as diagnostic information. This process is defined as a learning process. Moreover, since the parameter memorize | stored in the learning part 6 as diagnostic information can be updated or replaced | exchanged, the versatility of the electric equipment diagnostic system 1 can be improved.

  Further, when the diagnosis unit 7 diagnoses the winding unit 2 in a state where the electrical equipment is stopped, the feature amount detection unit 4 detects when the impulse having the same characteristic as the predetermined impulse is applied to the winding unit 2. The probabilities (likelihoods) of inputting the signals after the various feature values are converted into digital signals by the A / D conversion circuit 5 and generating waveform patterns of the respective feature values are stored in the learning unit 6 in advance. Is calculated based on each of the parameters, the parameter with the maximum generation probability (likelihood) is determined, and the winding unit 2 is diagnosed based on the diagnostic information corresponding to the parameter. If the failure is degraded, the degree of the failure is specified.

  Note that, as described above, the characteristic amount of the winding unit 2 includes voltage, current, and the like, and the frequency spectrum of these characteristic amounts and the time series of the frequency spectrum are also included as characteristic amounts. A plurality of feature amounts may be combined and used as a feature amount. Furthermore, a multidimensional distribution created from these feature quantities or a time series of multidimensional distributions may be used. Further, a feature amount obtained by reducing a multi-feature amount (multi-dimension) to one dimension or the like by using a principal component analysis or a self-organizing map may be used. In some cases, it may be difficult to detect a feature amount in any abnormal state of the winding portion 2. In that case, a waveform pattern of a feature amount in an abnormal state created on a computer by simulation or the like may be used. It is also possible to prepare a waveform pattern of feature amounts in a typical abnormal state. Thus, it is not necessary to prepare waveform patterns of feature amounts for all abnormal states, apply the hidden Markov model, parameterize them, and store them in the learning unit 6.

  In the learning unit 6, when the waveform pattern of the feature value is parameterized as described above, calculation is performed by applying a calculation algorithm including a forward algorithm and a backward algorithm. In the diagnosis unit 7, when the probability (likelihood) of generating a feature quantity pattern is calculated using the parameters stored in the learning unit 6, the calculation includes a forward algorithm or a backward algorithm. Apply algorithm to calculate.

  The display unit 8 displays a diagnosis result of the winding unit 2 of the electrical equipment diagnosed by the diagnosis unit 7 and is a display unit of a computer constituting the learning unit 6 and the diagnosis unit 7. The state of the winding unit 2 diagnosed by the diagnosis unit 7 is displayed on the computer display as described above, and can be transmitted to other devices by using communication means.

  The A / D conversion circuit 5, the learning unit 6, and the diagnosis unit 7 described above are collectively used as a hidden Markov model unit (HMM unit) 10 in the following description.

  FIG. 3 shows the same characteristics as the predetermined impulse used in the learning process described above when diagnosing the winding part 2 of the electric motor, for example, based on the parameters of the diagnostic information already stored in the learning part 6. FIG. 6 is a diagnostic explanatory diagram for diagnosing the winding part 2 by applying the impulse of FIG.

  The impulse generating circuit 3 shown in FIG. 3 applies an impulse having the same characteristics as the predetermined impulse used in the learning process described above when diagnosing the winding portion 2 to the winding portion 2. On the other hand, when a predetermined impulse is generated from the impulse generating circuit 3, the voltage across the winding portion 2 is detected by using, for example, the high voltage probe 11. Further, a current flowing through the winding unit 2 when a predetermined impulse is applied to the winding unit 2 is detected by a current sensor 12 set between the impulse generating circuit 3 and the winding unit 2. The high-voltage probe 11 and the current sensor 12 correspond to the feature amount detection unit 4 shown in FIG.

  FIG. 4 is a waveform diagram showing that the voltage detected by the high voltage probe 11 becomes a damped oscillation voltage waveform when the impulse output from the impulse generating circuit 3 is applied to the winding part 2. Since this damped oscillating voltage waveform is a characteristic waveform according to the characteristics and state of the winding section 2, the voltage signal detected by the high voltage probe 11 is converted into a digital signal by the A / D conversion circuit 5. The signal is used as a feature value, and the feature value is output to the diagnosis unit 7 of the hidden Markov model unit (HMM unit) 10. The diagnosis unit 7 recognizes the waveform pattern of the feature value, and calculates the probability (likelihood) of generating the waveform pattern of the feature value based on the respective parameters stored in advance in the learning unit 6 described above. Then, after determining a parameter that maximizes the generation probability (likelihood), the winding unit 2 is diagnosed based on diagnostic information corresponding to the parameter. Thus, if the winding unit 2 is diagnosed as normal, the normal state is diagnosed, and if it is diagnosed as a failure, the failure is specified. To display.

  In addition, since the current waveform of the current detected by the current sensor 12 set between the impulse generator 3 and the winding unit 2 is also a damped oscillation current waveform similar to the voltage, the detected current signal is expressed as A / The signal converted into a digital signal by the D conversion circuit 5 is used as a feature value, and the feature value is output to the diagnosis unit 7 of the hidden Markov model unit (HMM unit) 10. The diagnosis unit 7 recognizes the waveform pattern of the feature value, and calculates the probability (likelihood) of generating the waveform pattern of the feature value based on the respective parameters stored in advance in the learning unit 6 described above. Then, after determining a parameter that maximizes the generation probability (likelihood), the winding unit 2 is diagnosed based on diagnostic information corresponding to the parameter. Thus, if the winding unit 2 is diagnosed as normal, the normal state is diagnosed, and if it is diagnosed as a failure, the failure is specified. To display.

FIG. 5 shows that the back electromotive force is generated as an impulse by using the back electromotive force generating circuit 13 instead of the impulse generating circuit 3 shown in FIG. To do.
As shown in FIG. 5, the counter electromotive force generation circuit 13 includes a DC power supply 14 for passing a DC current through the winding portion 2, a switch 15 connected in series to the DC power supply 14, and the switch 15 is turned on. And a capacitor 16 charged in a state. When the switch 15 is turned on in a state where the winding portion 2 is connected in parallel to the capacitor 16, a direct current is passed through the winding portion 2. Therefore, when the switch 15 is turned off in this state, the winding portion 2 is turned on. A back electromotive force is generated in the line portion 2. Since this back electromotive force acts in the same manner as the impulse applied to the winding part 2, the detected back electromotive force is used as a feature amount, and the feature amount is used as the diagnosis unit 7 of the hidden Markov model unit (HMM unit) 10. To output. The diagnosis unit 7 recognizes the waveform pattern of the feature value, and calculates the probability (likelihood) of generating the waveform pattern of the feature value based on the respective parameters stored in advance in the learning unit 6 described above. Then, after determining a parameter that maximizes the generation probability (likelihood), the winding unit 2 is diagnosed based on diagnostic information corresponding to the parameter. Thus, if the winding unit 2 is diagnosed as normal, the normal state is diagnosed, and if it is diagnosed as a failure, the failure is specified. To display.

  Next, in a manufacturing factory for transformers and rotating machines, factory tests defined by standards such as JEC are performed. In this case, a lightning strike withstand voltage test with respect to an overvoltage at the time of a lightning strike or a cut-off withstand voltage test with respect to an overvoltage generated during a system cut-off operation is performed on a transformer or a rotating machine. In the lightning strike withstand voltage test, a lightning impulse withstand voltage test is performed, and in the breaking withstand voltage test, a switching impulse withstand voltage test is performed. The lightning impulse voltage used in the lightning impulse withstand voltage test has an impulse duration of several μs to several tens of μs, and the switching impulse voltage used in the switching impulse withstand voltage test has an impulse duration of several hundreds. μs to several ms. Further, the lightning impulse voltage and the switching impulse voltage each have a prescribed waveform, polarity, and peak value.

In the factory test as described above, the electrical equipment diagnosis system 1 of the present invention is used, and a lightning impulse voltage and a switching impulse voltage having a prescribed waveform, polarity, and peak value are converted into a diagnosis object such as a transformer or a rotating machine. If a hidden Markov model is applied to a waveform pattern such as voltage or current obtained by applying to the winding part, the winding part can be diagnosed as described above.
Note that the diagnosis can be performed on a single winding part before assembling a transformer, a rotating machine, or the like, or on a winding part after assembling.

  In the embodiment described above, the winding part of the transformer or the rotating machine is the object of diagnosis. However, any other electrical equipment can be the object of diagnosis as long as it has the winding part. Moreover, although the example of the voltage and the current is shown as the feature value, the sound or vibration generated when an impulse is applied to the winding portion may be used as the feature value. As described above, the waveform pattern recognition of the feature value is performed, and the probability (likelihood) of generating the waveform pattern of the feature value is calculated based on each parameter stored in advance in the learning unit 6 described above. When determining the parameter that maximizes the generation probability (likelihood) and diagnosing the winding unit 2 based on the diagnostic information corresponding to the parameter, even from the overall waveform shape of the waveform pattern of the feature amount Diagnosis is possible even from a partial waveform shape.

It is a conceptual explanatory view of a hidden Markov model. It is a systematic diagram of an electrical equipment diagnostic system. It is a voltage and electric current detection system diagram at the time of applying an impulse with respect to a coil part. FIG. 4 is a waveform diagram of a detection voltage in FIG. 3. A counter electromotive force generating circuit for generating a counter electromotive force in the winding portion. It is a wave form diagram for demonstrating the prior art.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Electrical equipment diagnostic system 2 Winding part 3 Impulse generation circuit 4 Feature-value detection part 5 A / D conversion circuit 6 Learning part 7 Diagnosis part 8 Display part 10 Hidden Markov model part

Claims (3)

An electrical equipment diagnosis system for diagnosing a winding portion constituting an electrical equipment,
Impulse generating means for generating a predetermined impulse corresponding to the characteristics of the winding part, and an impulse generated by the impulse generating means in each predetermined state of the winding part is applied to the winding part When the waveform pattern of the characteristic amount obtained in this case is parameterized based on the hidden Markov model, and each of the parameters is stored in the learning means, and when the state of the winding portion is diagnosed, the winding portion The calculation probability of the waveform pattern of the feature amount obtained when an impulse having the same characteristics as the impulse is applied is calculated based on the respective parameters stored in the learning means, and the parameter having the maximum generation probability is calculated. An electrical equipment diagnostic system comprising diagnostic means for diagnosing and judging the state of the winding portion based on the diagnostic means.   2. The electrical equipment diagnosis system according to claim 1, wherein a frequency for determining a state of the winding portion based on a waveform of the frequency spectrum of the feature amount or a time-series waveform of the frequency spectrum of the feature amount is set. An electrical equipment diagnosis system comprising an analysis means.   The electrical equipment diagnosis system according to claim 1 or 2, wherein the parameters stored in the learning means can be updated or exchanged.

Images