JP2018036113A - Signal processing device, and abnormality diagnostic device and abnormality diagnostic method of power apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technology that can easily diagnose internal abnormality or a deterioration state of a power apparatus without stopping an operating power apparatus.SOLUTION: A signal processing device according to the present invention is characterized to comprise acquisition means that continuously or intermittently acquires a signal being a function of time, frequency analysis means that computes a complex frequency spectrum of the acquired signal to calculate, time frequency analysis computing means that performs signal processing of the acquired signal from both of the time and frequency at the same time, noise component removal means that removes a noise component which is estimated to be included in the signal being the function of the time acquired, and display means that displays a source signal obtained after removing the noise component.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a signal processing device capable of diagnosing an internal abnormality or deterioration in an operating state of a power device, and a power device abnormality diagnosis device and abnormality diagnosis method using the signal processing device. More specifically, the present invention relates to a technique that can easily diagnose an internal abnormality or deterioration state of a transformer, which is an important device of electric power equipment, without stopping an operating transformer.

  A large number of power transformers are used in substations, transmission and distribution facilities, and various plants. When these power devices are damaged due to aging or various causes, partial discharge and abnormal noise are generated. These problems cause further deterioration of insulating materials and structural members used in the power equipment, and affect the life of the power equipment. For this reason, detecting the presence / absence of an abnormal signal of a power device is a very important technical problem in maintaining the reliability of the power device or monitoring / diagnosis of equipment.

Equipment monitoring / diagnosis methods include electrical methods for detecting pulse currents and high-frequency electromagnetic fields generated by partial discharge (see Patent Documents 1 and 2 and Non-Patent Documents 1 and 2), tank vibrations and sound waves in power equipment Physics and chemistry to estimate the degradation state from the analysis of the decomposition gas caused by partial discharge and arc discharge (DGA), oxidation degree, furfural analysis, polymerization degree and color change of insulating material, viscosity, etc. There is a special method.
In addition, recently, frequency analysis (FRA) for detecting a change in transfer function representing the frequency characteristics of the transformer winding has also been attempted.

JP 2009-216605 A JP 2012-042296 A

IEEJ Technical Report No. 1336, “Latest Trends in Transformer Abnormality Diagnosis Methods Using Electrical and Acoustic Methods” (March 2015) Power Transformer Reform Guidelines ", Electric Cooperative Research, Vol.65, No.1 (2009)

The problem to be solved by the present invention will be described by taking a transformer as an example of power equipment.
Transformer tank vibration is caused by electromagnetic force due to the load current flowing through the coil, magnetostriction due to the excitation current of the magnetic steel sheet constituting the iron core, magnetic attraction force at the joint of the iron core, magnetic attraction force against the tank due to leakage magnetic flux, or Maxwell stress. Or it is said to be caused by Lorentz force.
All of these forces are generated by a load current or an excitation current, and most of them generate a force proportional to the square of the current, but some of them do not exist like Maxwell's stress.
When any abnormality occurs in the transformer, for example, when the tightening force of the iron core is reduced or the coil is deformed, the tank vibration of the transformer also changes. Also, sound waves are generated by partial discharge and reach the tank wall. Furthermore, even when a physicochemical change occurs in the insulating oil or the insulator constituting the transformer, the deterioration gradually proceeds.

These abnormality or deterioration detection methods are roughly classified into electrical methods, acoustic methods, and physicochemical methods.
Although the method of detecting an electrical signal is excellent in terms of sensitivity, there is a problem that the actually acquired measurement signal has a poor SN ratio. This is because the partial discharge signal is weak and the phenomenon itself is irregularly distributed, and its handling often requires a statistical method, and it is derived from the power supply in the field where the transformer is installed. It can be mentioned that partial discharge measurement is often performed in an environment where external noise is superimposed.
In particular, in recent years, various power sources and manufacturing facilities using semiconductor devices have become widespread in addition to broadcast waves and communication radio waves, and these have become sources of noise. Therefore, it is necessary to pay attention to the discrimination from noise signals generated from these devices.
Further, when the cause of abnormality is only a mechanical factor and partial discharge does not occur, an essential problem remains that the abnormality cannot be detected from the electrical signal.
Furthermore, since the physicochemical method requires time for analysis, it has difficulty in time response, and is not suitable for measurement in real time.

  The present invention has been made in view of the above circumstances, and an object thereof is to provide a technique capable of easily diagnosing an internal abnormality or deterioration state of a power device without stopping the operating power device.

(1) The signal processing apparatus of the present invention includes an acquisition unit that continuously or intermittently acquires a signal that is a function of time, a frequency analysis unit that calculates and calculates a complex frequency spectrum of the acquired signal, Time frequency analysis calculation means for performing signal processing on the acquired signal from both time and frequency simultaneously, noise component removal means for removing a noise component estimated to be included in the signal as a function of the acquired time, and It is characterized by comprising display means for displaying the restored signal after removing the noise component.

(2) In the signal processing apparatus of the present invention, the means for calculating the complex frequency spectrum in the frequency analysis means is Fourier transform or fast Fourier transform, and a noise component is designated or estimated from the complex frequency spectrum obtained by the calculation. In addition, it is possible to provide an arithmetic device that performs an operation on a signal that is a function of a newly input time, while storing or learning a designated or estimated noise component.
(3) In the signal processing apparatus of the present invention, the means for performing signal processing in the time-frequency analysis calculation means is a wavelet transform with a plurality of levels of filter banks, and a spectrum corresponding to a noise component retained by the storage or learning is obtained. It is preferable to include a noise component removing unit that removes a noise component from a signal having a function of time by using a filter that can be removed or a basis function corresponding to the filter.

(4) In the signal processing device of the present invention, it is preferable that the device for calculating the complex frequency spectrum is configured using one or a plurality of mixers.
(5) The signal processing apparatus of the present invention includes a delay trigger circuit that generates a delay trigger for an input signal, and a basis function generator that generates a basis function signal using a trigger pulse of the delay trigger circuit. The signal of the basis function generator is mixed by a mixer, and the output of the mixer is amplified by an amplifier, and a time frequency analysis is performed by taking out the signal through a plurality of filters having different frequency characteristics. It is preferable to have a waveform synthesizing circuit that selectively extracts the output of the output from the switch circuit controlled in advance and restores the waveform based on the output extracted from the switch circuit.

(6) The signal processing apparatus of the present invention preferably includes an abnormal signal detection means for detecting an abnormal signal included in the original signal that is a function of time from the signal from which the noise component has been removed.
(7) The abnormality diagnosis device for a power device according to the present invention detects a vibration generated by the power device as a signal that is a function of time as described above, and removes a noise component included in the vibration. It is characterized by the ability to detect abnormalities in electric power equipment from the vibrations.

(8) The abnormality diagnosis method for a power device according to the present invention acquires a signal that is a function of time continuously or intermittently, calculates a complex frequency spectrum of the acquired signal by Fourier transform, The acquired signal is subjected to signal processing by wavelet transform simultaneously from both the time and frequency, the noise component estimated to be included in the signal having the acquired time as a function is removed, and the signal after the noise component is removed According to this signal, an abnormality of the power device is diagnosed.
(9) In the power equipment abnormality diagnosis method of the present invention, a noise component is specified or estimated from the complex frequency spectrum obtained by the Fourier transform, and the specified or estimated noise component is newly input while being stored or learned. It is preferable to perform an operation on a signal that is a function of time.

(10) In the abnormality diagnosis method for power equipment according to the present invention, the means for performing signal processing in the time-frequency analysis calculation means is a wavelet transform with a plurality of levels of filter banks, and corresponds to a noise component held by the storage or learning. It is preferable to remove a noise component from a signal that is a function of time by using a filter that can remove the spectrum of or a basis function corresponding thereto.
(11) In the power equipment abnormality diagnosis method of the present invention, a delay trigger is generated for an input signal, a basis function signal is generated by a trigger pulse of the delay trigger circuit, and the input signal and the signal of the basis function generator are Mixing with a mixer, and taking out the output signal of the mixer through a plurality of filters having different frequency characteristics to perform time-frequency analysis, and further selecting the output of each filter selectively with a pre-controlled switch circuit The waveform is preferably restored based on the output taken out from the switch circuit.
(12) In the abnormality diagnosis method for a power device according to the present invention, it is preferable to include means for monitoring the extracted output signal and issuing an alarm when an abnormality occurs.

According to the signal processing apparatus of the present invention, in addition to the electrical method (partial discharge measurement and frequency response analysis: FRA) and the physicochemical method, whether or not mechanical abnormality or partial discharge has occurred is detected. It is possible to provide a diagnostic device that can automatically detect and comprehensively determine the cause.
In particular, the signal processing device of the present invention has a great advantage that it can always monitor in the operating state of the power equipment.
Furthermore, according to the signal processing apparatus of the present invention, by incorporating the frequency analysis means, the time frequency analysis calculation means, and the noise component removal means, it is possible to remove the induction noise or inverter noise from the power source, and the SN ratio. Real-time measurement is possible while improving. Further, according to the present invention, it may be possible to obtain a result that is comparable to the electrical method in terms of detection sensitivity or resolution. Furthermore, according to the signal processing device of the present invention, it is possible to provide a signal processing device that is not easily affected by fluctuations in the power supply frequency.

In particular, the inverter noise cannot be removed by signal processing by Fourier analysis, but the frequency of the inverter noise and the tank vibration is separated from each other. By performing signal processing using wavelet transform using this fact, it is possible to remove inverter noise and to extract only signals from operating power equipment.
For these reasons, it is useful to apply them to transformers as power equipment, and use Fourier transform for frequency analysis means and wavelet transform for time frequency analysis calculation means in transformer tank vibration analysis. In the present invention, a signal processing apparatus that can realize these can be configured.
Various basis functions have been proposed for use in analysis by wavelet transform, but an odd function is advantageous because there is no DC component superposition.
Of course, the signal processing apparatus of the present invention can independently operate the frequency analysis means as a frequency spectrum analyzer and the time-frequency analysis calculation means as a time-frequency analyzer. Furthermore, although the present invention has been described with respect to a mechanical signal, it can also be used in combination with a partial discharge sensor signal.

1 is a circuit diagram showing a basic configuration of a signal processing apparatus according to the present invention. FIG. 2 is a circuit diagram showing a basic configuration of a frequency vector type spectrum analyzer provided in the signal processing apparatus shown in FIG. The phase relationship figure of the same frequency vector system spectrum analyzer. FIG. 2 is a circuit diagram showing a basic configuration of a wavelet analyzer provided in the signal processing apparatus shown in FIG. Explanatory drawing which shows the timing of the delay trigger pulse for time frequency analysis. Explanatory drawing which shows about the division | segmentation and synthesis | combination of a waveform using a multi-rate filter bank. The whole apparatus block diagram for acquiring the signal required for a diagnosis. The graph which shows the output waveform (CH1-CH3) and load current waveform (CH4) of the acceleration sensor measured from the transformer in operation. The graph which shows an example of the Fourier spectrum of the transformer tank vibration by a fast Fourier analysis (FFT analysis). The graph which shows an example of the Fourier spectrum of the harmonic component to the 30th order with respect to a power supply frequency. A second harmonic spectrum obtained from a transformer operating at a commercial frequency and an example of phase measurement, (a) is a graph showing in-phase component a ₂ , quadrature component b ₂ and absolute value C ₂ , (B) is a graph showing a time-series change of the phase angle θ ₂ . The in-phase component and the quadrature component of the third harmonic spectrum obtained from a transformer operating at commercial frequency, and a measurement example of the time series and phase of the absolute value are shown. (A) shows the in-phase component a ₃ and the quadrature component. component b ₃ and a graph showing the absolute value C _3, (b) is a graph showing a time series change of the phase angle theta _3. The in-phase component and the quadrature component of the fourth harmonic spectrum obtained from the transformer operating at the commercial frequency and the time series change of the absolute value are shown. (A) shows the in-phase component a ₄ , the quadrature component b ₄ and graph showing the absolute value C _4, (b) is a graph showing a time series change of the phase angle theta _4. Graph showing the spectral phase component _{A 1} and the quadrature component _{B 1} and the absolute value _{C 1} for the fundamental frequency 142Hz. Graph showing the phase components _{A 2} and the quadrature component _{B 2} and magnitude spectrum _{C 2} of the second harmonic spectrum with respect to the fundamental frequency 142Hz. The time-series changes of the in-phase component A ₃ and the quadrature component B ₃ of the third harmonic spectrum of the fundamental frequency 142 Hz and the absolute value spectrum C ₃ and the phase angle θ ₃ are shown. (A) is orthogonal to the in-phase component A _3. component B ₃ and a graph showing the magnitude spectrum C _3, (b) is a graph showing a time series change of the phase angle theta _3. Shows a chronological change of the second harmonic magnitude and phase and quadrature components of the spectrum of a typical fundamental frequency 332Hz than commercial frequency, (a) shows the phase component A ₂ and the quadrature component B ₂ and absolute graph showing the value spectrum C _2, (b) is a graph showing a time series change in the absolute value. The graph which shows an example of the Gaussian function applied as a basis function used for the analysis by wavelet transform. It shows an example of a result when wavelet analysis is performed on a transformer in operation, (a) is a graph showing a waveform when j = 1, (b) is a graph showing a waveform when j = 2, (C) is a graph showing the waveform when j = 3, (d) is a graph showing the waveform when j = 4, (e) is a graph showing the waveform when j = 5, (f) is j (G) is a graph showing a waveform when j = 7, and (h) is a graph showing a waveform when j = 8. Explanatory drawing which shows the display by a radar chart, and the setting of a determination level.

<First Embodiment>
Hereinafter, the case of analyzing the tank vibration of an oil-filled transformer will be described with reference to the drawings for the signal processing device according to the present invention.
FIG. 1 shows a basic configuration of a signal processing apparatus A according to the present invention.
An input signal necessary for vibration analysis detected by a sensor to be described later is amplified by the amplifier 1 and then input to a frequency vector analyzer (frequency analysis means) 2 and a time frequency analyzer (time frequency analysis calculation means) 3. The frequency vector analyzer 2 and the time frequency analyzer 3 are connected to the control device 4. In FIG. 1, reference numeral 5 indicates a display device (display means) connected to the arithmetic device 4.
In the first embodiment, the signal processing device A is configured by the amplifier 1, the frequency vector analyzer 2, the time frequency analyzer 3, and the arithmetic device 4. The amplifier 1 is not essential, and an amplifier may be provided on the sensor side that outputs a signal.

In the signal processing apparatus A of the present embodiment, the frequency analysis means (vector type spectrum analyzer; frequency vector analyzer) 2 performs Fourier transform, and the time frequency analyzer (wavelet analyzer) 3 performs wavelet transform.
The arithmetic unit 4 has a function of detecting a noise component from the Fourier spectrum obtained by the Fourier transform, storing and holding this information, and transferring this information to the time-frequency analyzer 3. The time-frequency analyzer 3 has a function of removing noise components using a plurality of filters described later based on the transferred information, and outputs a signal from which noise has been removed to the arithmetic unit 4. Further, the arithmetic device 4 has a function (abnormal signal detection means) for determining the presence or absence of an abnormal signal of a device that has extracted a signal based on the signal output from which the noise component has been removed. It has a function of outputting information for which presence or absence is determined to the display device 5.

The arithmetic device 4 incorporates a function for computation and signal generation, and an algorithm for diagnosing abnormality / deterioration by comparing time series data with each other.
As an example, the arithmetic device 4 includes a personal computer having a CPU (Central Processing Unit) and a storage unit, and a program (not shown) for realizing the above functions is loaded from the storage unit to the memory and executed. it can. As the storage unit, a nonvolatile memory such as a magnetic disk device or a flash memory, a volatile memory such as a RAM, or a combination thereof can be applied. An input device such as a keyboard or a mouse (not shown) may be connected to the arithmetic device 4 as a peripheral device.
Further, the storage unit of the arithmetic device 4 may be built in the arithmetic device 4 itself, or in another device such as a database server, and the arithmetic device 4 accesses these storage units by communication or the like. Thus, a configuration for executing processing and the like may be used.
The computing device 4 is not limited to a personal computer, and may be dedicated hardware in which the ability to perform the above functions is mounted on a memory chip or a dedicated element.

  Hereinafter, Fourier transform and wavelet transform necessary for vibration analysis will be described using an example of the signal as vibration information. If the period of the signal is T, the Fourier series is expressed by the following equation (1).

(1) the coefficients _{a 0, a n, b n} , c n in formula is calculated by the following equation (2) to (6) below. In the equation (6), θ _n represents a phase compensation angle.

  Further, an inverse Fourier transform, which is a transform from the frequency domain to the time domain, is defined by the following equation (7).

  The periodic function is expressed by the following equation (8).

The fast Fourier transform algorithm was devised by Cooley and Tukey in 1965 and is widely used in various fields today, but this calculation algorithm is also effectively used in this embodiment. Since 1982, wavelet analysis, which is time-frequency analysis, has been developed as a method for supplementing Fourier analysis and has been used in various fields. This calculation algorithm is also effectively used in this embodiment.
The wavelet analysis begins with an analysis using “Wavelets of constant shape” conceived by French oil technologist Morlet in the early 1980s. This method has an advantage that frequency analysis and extraction of discontinuous signals can be performed by selectively extracting a specific time domain. The wavelet transform Wψ (a, b) for the signal f (t) is expressed by equation (9).

  In Equation (9), ψ is a basis function, and here, explanation will be made using a Gaussian function of the following Equation (10). In the equation (10), since the period T is shifted by −b by the delay trigger circuit as shown in an embodiment described later, this means that the time b is shifted as a delay amount.

In each equation, the variable a is a scale parameter. If this value is reduced, observation takes a short time, and the frequency automatically spreads and high frequency can be observed. A variable b is a shift parameter for performing parallel movement of the waveform on the time axis.
FIG. 18 shows a Gaussian function normalized by (tb) / a and its first derivative function. Assuming that one waveform is approximately ± 3, the period of the waveform shown in FIG. 18 is 6a.
In the following description, the coefficient based on the commercial frequency power supply is a _n, b _n, shown in lowercase letters as c _n, coefficients based on the period other than the commercial frequency A _n, B _n, as C _n Expressed in uppercase alphabetic characters.

In the first embodiment of the signal processing apparatus A described above, the frequency analysis means (vector system spectrum analyzer) 2 can adopt the configuration shown in FIG. 2 as an example. An algorithm for designating or determining a noise component from the spectrum obtained by the vector type spectrum analyzer 2 is provided in the arithmetic unit 4.
The vector type spectrum analyzer 2 shown in FIG. 2 has two lines branched from the input part of the signal, the first mixer 10 and the filter 11 are incorporated in one line, and the second mixing is incorporated in the other line. A vessel 12 and a filter 13 are incorporated. One output unit of the local oscillator 14 is connected to the first mixer 10, the other output unit of the local oscillator 14 is connected to the second mixer 12 via the delay circuit 16, and the input unit of the local oscillator 14 is connected. The frequency sweep oscillator 15 is connected to the frequency sweep oscillator 15, and a control signal is supplied to the frequency sweep oscillator 15 from the arithmetic unit 4.

In FIG. 2, the first mixer 10 detects a component in phase with the cosine (or sine) included in the signal (hereinafter referred to as an in-phase component I). The second mixer 12 is configured to detect a component Q other than the in-phase component (hereinafter referred to as a non-in-phase component Q). Specifically, the cosine wave cos (2πf _LO t) transmitted from the local oscillator 14 that transmits the frequency specified by the frequency sweeping oscillator 15 is added to the first mixer 10, and the angular frequency πf _LO t A sine wave sin (2πf _LO t + θ) obtained by adding a phase compensation angle θ to the second mixer 2 is added to the second mixer 2 via the delay circuit 16 and is multiplied by a signal.
In FIG. 2, the detection result of the in-phase component is denoted as I (f _IF ), and the detection result of the non-in-phase component is denoted as Q (f _IF ).
If the phase compensation angle θ = 0, the configuration is the same as that of a general vector type spectrum analyzer. However, in this embodiment, the phase compensation angle θ designated by the arithmetic unit 4 is added to the delay circuit 16 so as to compensate the phase compensation. One feature of the embodiment shown in FIG. 2 is that a sine wave with the angle θ added is added to the second mixer 12.

This operating principle is shown in FIG.
It is assumed that the spectrum size is originally a value represented by the following expression (11).

  Here, let us consider a case where the asynchronous component Q is attenuated to (Q−ΔQ) as shown in FIG. In this case, the magnitude of the spectrum can be expressed as a change in the non-orthogonal component without changing the magnitude of the frequency spectrum as shown in the following equation (12) by introducing the phase compensation angle θ.

  Further, the relationship of the following expression (13) or (14) is established between ΔQ and sin.

This signal processing is an effective means for visualizing a spectrum change when attention is paid to a specific frequency spectrum and the spectrum size is normalized and expressed.
In addition, the vector system spectrum analyzer 2 can remove the noise component from the obtained spectrum by providing the arithmetic device 4 with a storage device or software for designating or determining the noise component.

In the signal processing apparatus A of the first embodiment, the time-frequency analyzer 3 is constituted by a wavelet analyzer shown in FIG. 4 as an example.
In the time-frequency analyzer 3 of this embodiment, an amplifier 20, a third mixer 21 and an amplifier 22 are incorporated in a line from the input unit, and a delay trigger circuit is provided in a line between the amplifier 20 and the third mixer 21. 23 is connected, and the output unit of the basis function generator 24 is connected to the third mixer 21. The arithmetic device 4 is connected to the input portion of the delay trigger circuit 23 and the input portion of the basis function generator 24, and a control signal is supplied from the arithmetic device 4 to the delay trigger circuit 23 and the basis function generator 24.
In the basis function generator 24, parameters such as a function and a scale parameter a are set in advance by the arithmetic unit 4. The shift parameter b is given from the delay trigger circuit 23 to the basis function generator 24. As an example, as shown in FIG. 5, delay trigger pulses of an integral multiple (i times) of the time width obtained by dividing the power supply cycle T into N are sequentially generated from the reference point of the power supply frequency.
This relationship is shown in the following equation (15).

The delayed trigger pulse generated here is input to the basis function generator 24, a basis function necessary for wavelet transform is generated, and is supplied to the third mixer 21.
The delay trigger circuit 23 has a convolution integration function in wavelet calculation, and the basis function generator 24 generates a basis function necessary for wavelet transformation, and the scale parameter a can be set.

The output of the third mixer 21 is amplified by the amplifier 22 and then input to a plurality of filters 25 having different frequency characteristics in conjunction with the scale parameter a. Each filter 25 is connected to a switch circuit 26 controlled by the arithmetic unit 4, and is connected to a waveform synthesis circuit 27 that outputs a synthesized output from the switch circuit 26 connected to each filter 25.
As an example, the waveform synthesis circuit 27 can reproduce the original waveform from which noise has been removed by the wavelet algorithm, using the multi-rate filter bank shown in FIG. The basis function used in the waveform synthesis circuit 27 is preferably an odd function.
Here, the mother wavelet as a basis function will be briefly described. It is created using a hierarchical structure of function space called multiresolution approximation (MRA). When the sequence is {p _k }, when the function φ (x) satisfies the relationship of the following equation (16), this is called a two-scale relationship.

A function that satisfies this relationship is called a scaling function. Using the scaling function and the new number sequence {q _k }, the function of the following equation (17) is defined.

This is called the mother wavelet and satisfies the basis function condition.
In the wavelet transform of the equation (9), the time and frequency coordinates (b, 1 / a) can be discretized into the following equation (18) by using two integers j and k.

Here, if the wavelet transform Wψx (a, b) for the signal x (t) is expressed as d _k ^(j) , it can be expressed by the following equation (19).

  In this case, the inverse transformation is given by the following equation (20).

  Furthermore, if one of the double sums on the right side of the equation (20) is defined as the following equation (21), the following equation (22) is obtained.

  Here, a function represented by the following equation (23) is introduced.

In this equation (23), j is called a level. This equation (23) can be rewritten in a recursive form for x _j (t) as in the following equation (24).

  Here, for example, there is a relationship of the following expression (25).

In the formula (25), x ₋₁ can be further decomposed into the following formula (26).

Thus, it can be seen that the level can be lowered by 1 using the function x _j (t), and the resolution is halved each time.
The function x _j (t) can be expressed by a linear combination as shown in the following equation (27) using the scaling function φ (t).

In equation (27), the scaling function itself on the right side is the same regardless of the level. The mother wavelet is also synthesized from the scaling function.
Here, waveform division and synthesis using the multi-rate filter bank shown in FIG. 6 will be described.
For example, individual H ₀ (Z) and H ₁ (Z) on the split side shown in FIG. _6A correspond to the relationship between the above x _j (t) and the following equation (28).

  Further, (↓ 2) shown in FIG. 6 means that the frequency is divided into a high frequency component and a low frequency component. This is called band-division subband decomposition. Further, “Ditail” means that the decomposition is further advanced to a lower level, and “Approximation” means that the decomposition is stopped as a reasonable approximation.

In the case of the synthesis side shown in FIG. 6B, F ₀ (Z) and F ₁ (Z) are functions used for waveform synthesis, and (↑ 2) means that frequencies are synthesized. .
The division side shown in FIG. 6A corresponds to the filters 25-1 to 25-n in FIG. 4, and the synthesis side in FIG. 6B corresponds to the waveform synthesis circuit 27 in FIG. .

The arithmetic device 4 incorporates functions of control, computation, and signal generation, and software for diagnosing abnormality / deterioration by comparing time series data with each other. In addition, monitoring software or functional elements are incorporated so that the phase compensation angle θ and the waveform display after noise removal can be monitored over time, and the display device 5 can display the above-described change over time. Have.
In this embodiment, a noise component removing means 28 is constituted by the arithmetic device 4, the mixer 21, the delay trigger circuit 23, the basis function generator 24, the plurality of filters 25, the switch circuit 26 and the waveform synthesizer 27.

In general, noise is an unnecessary high-frequency component included in a signal. Conversely, if a signal includes some sudden abnormal component, the abnormality can be detected by detecting the high-frequency component.
For example, this can be realized by using Daubechies N = 8 as a basis function. It is also possible to obtain displacement, speed, and acceleration by calculation and find out abnormal values from these. In this case, it is convenient to use a spline function.
The display method includes a Fourier spectrum display to be described later based on FIG. 10, a display by a time series change diagram of a complex frequency vector and a phase to be described later based on FIGS. 11 to 17, a synthesized waveform by the following equation (29), and a diagram. There are a waveform display and determination level setting by a radar chart in units of one cycle shown in FIG. However, in Equation (29), An and Bn are treated as 0 for noise components.
In addition, by constantly monitoring the radar chart in units of one cycle shown in FIG. 20, it is possible to determine the presence or absence of an abnormality from a sudden change in the pattern or signal magnitude.

Hereinafter, the present invention will be described in more detail by taking as an example a case where the above-described signal processing apparatus A is used and the vibration of the power transformer is handled as an analysis signal.
FIG. 7 is a configuration diagram of a diagnostic device for obtaining signals necessary for diagnosis of the power transformer.
One or a plurality of vibration sensors (signal acquisition means) 32 are attached to the tank wall surface of the power transformer 31, and the output part of the vibration sensor 32 is connected to the input part of the signal processing apparatus A according to the present embodiment. . The vibration and acceleration of the tank wall surface detected by these vibration sensors 32 are input to the signal processing apparatus A as input signals. In this example, the vibration sensor 32 is used as acquisition means for acquiring a signal that is a function of time continuously or intermittently. However, instead of the vibration sensor 32, an acceleration sensor may be used.

Moreover, since the vibration of the tank wall surface moves in conjunction with the power source of the power transformer, it is preferable that the voltage waveform 33 and the current waveform 34 are taken in as an input signal of the signal processing apparatus A as necessary. In this state, in addition to steady tank vibration, an abnormality signal generated due to abnormality or deterioration inside the tank is included. Furthermore, noise derived from the measurement line or the power source of the measuring instrument is mixed.
The transformer's tank contains a primary coil and a secondary coil consisting of an iron core and windings, and the tank is basically based on the power supply frequency while the primary coil and secondary coil are energized. Vibrate.

  The outline of the transformers to be measured is shown in Table 1 below. The load is a pump inverter drive motor with a maximum output of 750 kW and 350 kW, and one or a plurality of pumps are always operated. The transformer used in this example is a transformer that has actually been continuously operated since its operation started in 1975, and is a 41st year transformer. The general shape of this transformer is a transformer having a rectangular steel plate tank with a height of about 2 m, a width of 800 mm, and a depth of 400 mm, and a coil consisting of a core yoke, a primary winding and a secondary winding. It is a transformer of the type in which rectangular frame type reinforcing stays are provided at two positions that divide the height of the tank peripheral wall into three equal parts.

The memory length of the digital oscilloscope used for waveform measurement is 1 M sample. When the sampling time is 10 μs, waveform data for 10 seconds, that is, 60 Hz × 10 seconds = 600 cycles can be acquired in one measurement.
FIG. 8 shows an example in which the tank vibrations at the three side surfaces of the transformer tank are actually measured by a vibration sensor (vibration sensor GH-313A manufactured by Keyence) (CH1 to CH3: numbers 1 to 3 in FIG. 8). Notation). The waveform at the bottom (CH4: indicated by the numeral 4 in FIG. 8) is a signal obtained by measuring the signal from the current transformer with the clamp CT. In FIG. 8, the vertical axis indicates the signal intensity of the vibration sensor, and the horizontal axis indicates the time corresponding to one cycle of the commercial frequency.
It can be seen that inverter noise is superimposed on the commercial frequency current in the previous transformer, and that the tank vibration of the transformer has a much more complicated vibration waveform compared to the commercial frequency. It can be seen that the waveform size and phase are different.
This is because the transformer used for the measurement is a three-phase device, the instantaneous current flowing through the coil of the transformer varies from phase to phase, the induced electromagnetic force is a combination of these forces, and the vibration sensor is at the signal level. It is estimated that the main cause is that it is small and susceptible to external noise, particularly inverter noise.

FIG. 9 shows an example in which the Fourier spectrum of the vibration sensor attached to the longitudinal wall surface of the rectangular parallelepiped tank is analyzed by Equation (5) in order to confirm the frequency component of the tank vibration of the transformer used for the measurement. In FIG. 9, the vertical axis represents the Fourier spectrum intensity, and the horizontal axis represents the order of the harmonics with respect to the commercial frequency.
From FIG. 9, there is a wide frequency spectrum having a peak at around 3 kHz in addition to the signal component derived from the power supply frequency, that is, a harmonic of an integral multiple of 60 Hz. The inventor has confirmed by separate analysis that this frequency spectrum is considered to be derived from the switching frequency of the PWM inverter.
Therefore, the frequency characteristics of the vibration sensor are also taken into consideration, and the following analysis will be limited to frequencies below this.

Further, in the Fourier spectrum of FIG. 9, there are spectrums that are estimated to be causes other than the high frequency component of the commercial frequency and the inverter, for example, 142 Hz, 217 Hz, 334 Hz, 490 Hz, 693 Hz, and 713 Hz.
FIG. 10 shows the Fourier spectrum up to the 30th harmonic (1800 Hz) of the commercial frequency, using measurement data for 600 cycles obtained from the vibration sensor attached to the short wall surface of the rectangular parallelepiped tank of the transformer. Is. In FIG. 10, the vertical axis represents the Fourier spectrum, and the horizontal axis represents the order of the harmonic with respect to the commercial frequency of 60 Hz.
Although a slight DC component (a ₀ ) is superimposed on this Fourier spectrum, in this example, the spectrum of the frequency 240 Hz (the signal at the position indicated by numeral 4 in FIG. 10) is maximized, and the even-order signal is dominant. It can be seen that it is.
In addition, the spectrum is less attenuated with respect to even-order harmonics. This means that the transformer tank vibration is caused by electromagnetic force proportional to the square of the current.
On the other hand, in the Fourier spectrum shown in FIG. 10, an increase in the odd-order component (spectrum at the position indicated by the odd number in FIG. 10) is observed as the order increases. This is considered due to the influence of inverter noise as shown in FIG.
In such a case, these frequency spectra are handled as noise.

Next, the time series change of the phase of the Fourier spectrum will be described.
FIG. 11 shows the in-phase component of cos and the in-phase component of the frequency spectrum included in the waveform of 600 cycles in total for 10 seconds with respect to the second harmonic of 120 Hz, that is, the value of equation (3) and the in-phase component of sin (that is, orthogonal to cos). Component), that is, a value of the equation (4) (hereinafter referred to as an orthogonal component) is separated and expressed as a time series together with the spectrum size given by the equation (5). These values are in a relationship of being absolute values when squared and square roots are taken. In FIG. 11, the vertical axis represents the Fourier spectrum, and the horizontal axis represents the time for 600 cycles (10 seconds).
FIG. 11A shows that the absolute value of the spectrum is substantially constant, but the ratio of the in-phase component (cos and in-phase component) and the quadrature component (sin and in-phase component) changes in time series. Further, as shown in FIG. 11B, it can be seen that the phase compensation angle θ ₂ calculated by the equation (6) gradually changes. This is caused by fluctuations in the power supply frequency, and in the calculation, the calculation is performed with the commercial frequency fixed at 60 Hz.

FIG. 12 shows in-phase components and quadrature components for the third harmonic 180 Hz, which are shown as time series. In FIG. 12, the vertical axis represents the Fourier spectrum, and the horizontal axis represents the time for 600 cycles (10 seconds).
From FIG. 12, it was found that the absolute value of the spectrum is small compared to the odd-order components and thus has a large variation, but is almost constant for each waveform and the phase gradually changes.
FIG. 13 shows the in-phase component, the quadrature component, and the phase change with respect to the fourth harmonic 240 Hz having the largest spectrum as a time series. In FIG. 13, the vertical axis indicates the Fourier spectrum, and the horizontal axis indicates the time for 600 cycles (10 seconds).
FIG. 13 shows that the absolute value of the spectrum is almost constant, but the phase gradually changes.

In this way, the absolute value of the Fourier spectrum of the harmonic component of the power supply frequency is almost constant, and the fact that the phase gradually changes over time is a fact that has been found for the first time this time. It was confirmed that the properties are common to even-order harmonics.
Although not shown in the present specification, when the present inventor studied, the phenomenon that the absolute value of the Fourier spectrum of the harmonic component of the power supply frequency is almost constant and the phase gradually changes with time, We were able to confirm that it occurred at all harmonics with respect to the power supply frequency.
Both the in-phase component and the quadrature component are complex numbers, and when these are squared and the root is taken to obtain an absolute value, the constant value is a real number. Normally, only one point of FIGS. 11 to 13 is analyzed in the Fourier transform, but it has been found in this test that even if the same spectrum is seen in time series, it includes a variation.

This is considered to be because the power supply frequency is not exactly 60 Hz and there is a minute frequency fluctuation, so that the phase gradually changes. In this example, the frequency fluctuation for about 1 cycle is confirmed in 600 cycles (10 seconds).
As described above, it has been clarified that the Fourier analysis using the Fourier transform can detect not only the absolute value of the Fourier spectrum but also the minute frequency fluctuation through the phase change. On the other hand, it can be said that the Fourier analysis based on the Fourier transform algorithm with a fixed fundamental frequency is affected by such a small power supply frequency.

In such a case, the vector type spectrum analyzer 2 shown in FIG. 2 has an advantage that the frequency spectrum can be calculated without being affected by the frequency fluctuation because the calculation is performed based on the power supply frequency. Further, by analyzing the frequency spectrum, it is possible to identify external noise including power supply induction.
In the vibration of the tank, the magnitude of the spectrum derived from the power supply frequency does not change, but if the vibration is caused by something else, such as a coil or iron core, or something other than the power supply, the tank will While receiving vibration, the tank vibration has its spectrum attenuated in time.
If such a damping vibration and a power source are distinguished from each other in handling an algorithm in an electric signal and separated, it is possible to extract the original vibration of the tank.

  On the other hand, in the Fourier spectrum shown in FIG. 9, the spectrum exists at frequencies other than the higher harmonic components of the commercial frequency, for example, 142 Hz, 217 Hz, 332 Hz, 490 Hz, 693 Hz, and 713 Hz. This was confirmed by the hammering test conducted. The hammering test conducted here is the measurement of vibrations obtained by striking a plurality of locations on the side wall of the transformer tank used for measurement with a hammer, and a peak component is generated in the obtained vibration waveform. It is a test that measures whether or not.

The result of calculating the time-series change of the Fourier spectrum using the above frequency as the fundamental frequency is shown below.
FIG. 14 shows the frequency spectrum for the fundamental frequency 142 Hz, and FIG. 15 shows the frequency spectrum for the double frequency of the fundamental frequency 142 Hz. In FIG. 14, the vertical axis indicates the Fourier spectrum, and the horizontal axis indicates the time for 600 cycles (10 seconds).
It is clear that the magnitude of the spectrum is not constant for either frequency spectrum.
FIG. 16 shows the time-series changes in the in-phase component, quadrature component, absolute value, and phase compensation angle of the fundamental frequency 142 Hz, the third harmonic spectrum. In FIG. 16, the vertical axis represents the Fourier spectrum, and the horizontal axis represents the time for 600 cycles (10 seconds).
Also in this case, it can be seen that the size of the spectrum is not constant.
FIG. 17 shows the time-series changes of the fundamental frequency 332 Hz, the in-phase component, the quadrature component, the absolute value, and the phase compensation angle θ ₃ of the second harmonic spectrum. In FIG. 17, the vertical axis represents the Fourier spectrum, and the horizontal axis represents the time for 600 cycles (10 seconds).

Also in this case, it can be seen that the size of the spectrum is not constant.
From the examples shown in FIGS. 14 to 17, in the transformer used in the test, it is considered that the generation and attenuation of vibrations by some excitation source are repeated at frequencies other than the harmonics of the power supply frequency. found.
As is clear from the comparison of these analysis examples, the frequency spectrum due to the harmonic component that is an integral multiple of the power supply frequency has little variation over time, but the frequency spectrum due to the frequency that is not the harmonic component of the power supply frequency is It was found that time variation due to generation and decay was observed over time.

That is, it was found that by using the signal processing apparatus A according to the present invention, it is possible to make a diagnosis while discriminating between vibrations derived from the power source and mechanical vibrations caused by some other cause.
In other words, whether or not the vibration is synchronized with the frequency (cycle) derived from the power source is used as a criterion, so that a specific vibration frequency is related to the vibration of the device or the power frequency. It is possible to distinguish whether the vibration is independent or not, and if the vibration that does not depend on the power supply frequency shows time fluctuation due to attenuation over time, it can be diagnosed that an abnormality has occurred and can be used as a basis for diagnosing the power equipment. I found out.

Next, time frequency analysis using wavelet transform will be described.
When analyzing the vibration of a transformer, the wavelet transform can grasp both time and frequency. In the analysis using the above Fourier series, only the frequency can be grasped.
Wavelet transform can use filters with various frequencies, and can remove extra components for each frequency, so noise can be eliminated by reconstructing the waveform based on the remaining signal after removing extra components. Only the removed tank vibration can be taken out. If the tank vibration includes abnormal vibration, it can be diagnosed whether or not the transformer is abnormal.
In the time-frequency analysis using the wavelet transform of this embodiment, the Gaussian function of Expression (10) is used as the basis function. The form of the Gaussian function is illustrated in FIG.
Here, the scale parameter a was set as follows using the sampling time Ts of the oscilloscope as a guide.

  As described above, when the period of the Gaussian function is assumed to be 6a, the standard of the frequency region to be observed corresponding to the scale parameter a is shown in Table 2 below.

FIG. 19 shows the result of wavelet analysis for the vibration waveform shown in FIG. In FIG. 19, the vertical axis indicates the standard value of the signal intensity at each level, and the horizontal axis indicates the time corresponding to one cycle.
The numerical values of j = 1 to 8 in FIG. 19 are values of the scale parameter a corresponding to the level j in Table 2.
The magnitude of the spectrum is expressed by standardizing from the maximum value and the minimum value, and high frequency components having a strong spectrum appear in FIGS. 19 (a), 19 (b), and 19 (c).
This is presumed to be caused by the fact that inverter noise, which is external noise, is mixed in the measurement circuit as inductive noise.
In FIG. 19 (d) or FIG. 19 (e), the high frequency component disappears, and this is presumed to be a signal derived from tank vibration captured by the acceleration sensor. Thus, it has been shown that in wavelet analysis, a filter can be constructed using the difference between the frequency of a signal and noise, and inverter noise can be removed. Therefore, if the signals shown in FIGS. 19A to 19C determined to contain noise are removed and the signals shown in FIGS. It is possible to obtain a signal consisting only of the vibration of the transformer from which the noise is removed.

  That is, it has been found that when wavelet transform is performed, noise components can be removed from the filter bank using a switch circuit.

As described above, the existence of a frequency that does not depend on the power supply frequency can be regarded as a part of the vibration inherent in the mechanical system, and the frequency spectrum and the vibration from which noise is removed using the signal processing apparatus A according to the present invention. By analyzing the waveform, it is possible to determine whether the transformer is abnormal. That is, it is possible to diagnose an abnormality of a device accompanied by vibration such as a transformer.
In the previous embodiment, wavelet analysis is used by using a plurality of filters 25 shown in FIG. 4, and a waveform including high-frequency component noise is removed from waveforms obtained according to the value of the scale parameter. A waveform associated with the vibration originally generated by the transformer can be obtained.
The waveform excluding this noise is measured and compared with the waveform measured in advance with a transformer that does not have problems such as deterioration. If there is a difference, the measured primary coil or secondary of the transformer It can be determined that this is a transformer that may cause deterioration or problems in any part including the coil or the winding core. For this reason, a waveform measured in advance with a transformer that does not have a problem such as deterioration is input to the arithmetic device 4 and compared with the measured waveform excluding the previous noise. A function of detecting as a signal is provided.
With this function, the arithmetic device 4 functions as a means for detecting an abnormal signal of the transformer.

In addition, it is possible to reproduce a signal waveform with periodicity like regular tank vibration by inverse Fourier transform, but in an environment where the power supply frequency fluctuates, in the case of analysis with a fixed fundamental frequency, an error factor It becomes.
The present invention has such a feature that such an influence can be eliminated by using the wavelet transform using the multi-rate filter bank shown in FIG.

  A ... signal processing device, 1 ... amplifier, 2 ... frequency vector analyzer (frequency analysis means), 3 ... time frequency analyzer (time frequency analysis calculation means), 4 ... calculation device, 5 ... display device (display means), 10 ... 1st mixer, 11 ... filter, 12 ... 2nd mixer, 13 ... filter, 14 ... local oscillator, 15 ... frequency sweep oscillator, 16 ... delay circuit, 20 ... amplifier, 21 ... 3rd mixer , 22 ... amplifier, 23 ... delay trigger circuit, 24 ... basis function generator, 25 ... filter, 26 ... switch circuit, 27 ... waveform synthesis circuit, 28 ... noise component removing means, 31 ... transformer, 32 ... vibration sensor ( (Signal acquisition means), 33 ... voltage waveform, 34 ... current waveform.

Claims (11)

An acquisition means for continuously or intermittently acquiring a signal that is a function of time;
Frequency analysis means for calculating and calculating a complex frequency spectrum of the acquired signal;
A time-frequency analysis calculation means for simultaneously performing signal processing on the acquired signal from both sides of time and frequency;
Noise component removing means for removing a noise component estimated to be included in the signal having the acquired time as a function;
A signal processing apparatus comprising display means for displaying a restored signal after removing the noise component.   In the frequency analysis means, the means for calculating the complex frequency spectrum is Fourier transform or fast Fourier transform, and a noise component is designated or estimated from the complex frequency spectrum obtained by the calculation, and the designated or estimated noise component is stored or stored. The signal processing apparatus according to claim 1, further comprising an arithmetic unit that performs an operation on a signal that is a function of a newly input time while learning. The means for performing signal processing in the time frequency analysis calculation means is a wavelet transform with a plurality of levels of filter banks,
There is provided a noise component removing means for removing a noise component from a signal having a function of time by using a filter capable of removing a spectrum corresponding to the noise component held by the storage or learning or a basis function corresponding to the filter. The signal processing apparatus according to claim 2.   The signal processing apparatus according to any one of claims 1 to 3, wherein the apparatus for calculating the complex frequency spectrum is configured using one or a plurality of mixers.   A delay trigger circuit for generating a delay trigger for the input signal, and a basis function generator for generating a basis function signal by a trigger pulse of the delay trigger circuit, and the input signal and the signal of the basis function generator are mixed by a mixer Mix and mix the output of the mixer with an amplifier to extract the signals through multiple filters with different frequency characteristics, and then select the output of each filter with a pre-controlled switch circuit. 5. The signal processing apparatus according to claim 1, further comprising a waveform synthesis circuit that restores a waveform based on an output extracted from the switch circuit and based on an output extracted from the switch circuit.   6. The signal processing apparatus according to claim 5, further comprising: an abnormal signal detection unit that detects an abnormal signal included in an original signal that is a function of time from a signal from which a noise component has been removed.   As a signal that is a function of time according to claims 1 to 6, the apparatus has the ability to detect vibrations generated by the power equipment and detect abnormalities in the power equipment from the vibrations from which noise components included in the vibrations are removed. An apparatus for diagnosing abnormality of electric power equipment, characterized by   A signal that is a function of time is acquired continuously or intermittently, and the complex frequency spectrum of the acquired signal is calculated and calculated by Fourier transform, and the acquired signal is simultaneously converted from both time and frequency by wavelet transform. Signal processing is performed to remove a noise component estimated to be included in the signal having the acquired time as a function, to obtain a signal after the removal of the noise component, and to diagnose an abnormality of a power device according to this signal A method for diagnosing abnormalities in power equipment.   Designating or estimating a noise component from the complex frequency spectrum obtained by the Fourier transform, and performing computation on a signal that is a function of time newly input while storing or learning the designated or estimated noise component. The power equipment abnormality diagnosis method according to claim 8.   The means for performing signal processing in the time-frequency analysis calculation means is a wavelet transform with a plurality of levels of filter banks, and a filter capable of removing a spectrum corresponding to a noise component held by the storage or learning or a basis function corresponding thereto The method according to claim 8, wherein a noise component is removed from a signal having time as a function.   Generate a delay trigger for the input signal, generate a basis function signal with the trigger pulse of the delay trigger circuit, mix the input signal and the signal of the basis function generator with a mixer, and output the mixer output signal with frequency characteristics The time frequency analysis is performed by extracting through a plurality of different filters, and the output of each filter is selectively extracted by a switch circuit controlled in advance, and the waveform is based on the output extracted from the switch circuit. The method for diagnosing abnormality of a power device according to any one of claims 8 to 10, wherein the power is restored.

Images