JPS61721A - Abnormality detector - Google Patents

Abstract

PURPOSE:To enable detection of an abnormality point regardless of a DC component, by removing previously the DC component by a removing filter for uniformalization to the constant amplitude size and removing HF noise by an adaptability type of filter and judging by a differential filter an abnormality of wave form. CONSTITUTION:A DC removing filter 3 and an automatic gain controller 4 makes a mean value of amplitude of a measuring signal x constant and further, the control is made without a distorted wave form by the DC component. An adaptability filter 9 is composed of high-speed Haar convertor 6, step number counter 7 and high-speed inverted Haar convertor 8 for removal of HF noise. A disciminating apparatus 11 by employment of a differential signal Wk which is an output of a differential filter 10 and of the step number counter 7 detects a pulse-type signal showing an abnormality included in a counting signal x. By this arrangement, the pulse-type signal in the counting signal can rapidly be detected without being affected by fluctuation of the amplitudes of the counting signal and by the DC component.

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention relates to an abnormality detection device that detects an unusual signal that suddenly appears in a measurement signal based on measurement signals obtained from various sensors. . An example will be explained in which an abnormality in a device is detected from a signal from a vibration accelerometer attached to a lunar power generation device or the like.

[Prior art]

Conventional abnormality detection devices using measurement signals from vibration accelerometers are based on the principle of primarily determining the distribution of the vibration power spectrum and detecting the presence of abnormal frequency spectral components from this distribution. Normally, in order to obtain the distribution of this power spectrum, a method has been used in which an analog resonant circuit is provided and the resonant frequency is manually varied to measure the power of the resonant voltage. This method was well suited for determining whether an abnormality occurred in the equipment and finding the frequency of vibrations caused by the abnormality. However, since the abnormality is detected based on the frequency spectrum distribution of the measurement signal, it is difficult to detect the abnormality without delay.

As a means to improve this drawback, an automatic gain controller that adjusts the amplitude of the measurement signal based on the measurement signal for a certain period of time in the past is used to adjust the average value of the amplitude of the measurement signal to be constant, and this is adjusted to ) transform and inverse Haar transform, smoothing is performed using an adaptive low-pass filter configured to automatically change the passband according to the properties of the measured signal, and the difference between this smoothed value and the measured signal is calculated as the average value. Calculate the sample variance assuming zero noise.

Furthermore, the signal generated at the abnormal point is emphasized by differentiating the smoothed value through a differential filter, and the square of this differentiated signal is divided by the value of the sample variance to obtain the ratio. If this ratio is larger than the set constant, measurement is performed. There is a method of detecting an abnormality in the device by determining that there is a signal generated at an abnormal point in the signal, but the automatic gain adjuster calculates the average value of the amplitude of the measurement signal over a certain period of time in the past and the measurement signal. The amplitude of the measurement signal is adjusted by simply making it inversely proportional to the averaged value of the difference Ω.
If a DC component is superimposed on the measurement signal due to improper adjustment of the DC amplifier of the sensor, the amplitude of the DC component will also be adjusted and the magnitude will change, causing waveform distortion of the measurement signal and making it impossible to detect an abnormality. It was sometimes influenced by the DC component.

[Summary of the invention]

This invention provides a DC removal method that removes DC components by subtracting the average value of the measurement signal over a certain period of time from the measurement signal in a conventional abnormality detection device that includes an automatic gain adjuster, an adaptive filter, and a differential filter. Add a filter.

The purpose of this invention is to remove the DC component superimposed on the measurement signal and manually input it to the automatic gain adjuster, thereby eliminating the drawbacks of the conventional method and further improving the ability to detect abnormal points. Figure 1 depicts an example of the measurement signal of the vibration accelerometer attached to the power generation equipment. This is the waveform of the signal generated at the point. This invention attempts to detect a pulse-like signal (1) in a measurement signal without being influenced by fluctuations in the amplitude of the measurement signal and DC components and without delay.

[Embodiments of the invention]

FIG. 2 shows an embodiment of the present invention. Note that although the measurement signal can take the form of either an analog quantity or a digital quantity depending on the configuration of the sensor, an example in which a measurement signal of an analog quantity is input will be explained.

(2) in the figure shows the measurement signal X from the vibration accelerometer with a period Ts
It is an A/D converter that converts the digital signal
I-1)・The above digital signal Xk until time T8
s

(41 is an automatic gain cylinder, from the present to the past (J+L
-1)・The above DC removed signal Fk up to time T8,
l;"k-1,..., the above D based on Fk+x-x-L
The magnitude of the C removal signal pk is adjusted and output as an AGC signal Qk. However, J and L are positive integers and J<L. (
5) is the above AGC signal Qk every cycle Tll (N-1)
・It is a data buffer that stores while updating over the past up to Tll, and stores the current value Gk as FN, the past value by Tl as Fs-s, and in this way, (N-1)・Tl.
The past values are stored as FO and these are output. (
6) is a high-speed/・-Role converter which performs Haar transform on the stored values Fo, F■,・Fist・, FN-s in the data buffer (5) and converts them into Norse vectors CO, CI, .・、C
Output N-5.

However, N is a positive number given as a power of 2, N=2f1
) ・・・・・・・・・(1) (7) is a step number determiner, and the above-mentioned Norse vectors Co, CI are determined using the determined appropriate positive positive number Σ as an allowable value.

..., CI1, subsequence co, CL, ..., CM- of the Haar spectrum according to certain rules described below.
1 is selected and output, and an exponential plane representing the order h of the subsequence of the 2·-le spectrum at that time, that is, the adaptive order M is output. In this case, assuming that the adaptive order M is a power of 2, the distance between it and the exponential plane is M=2...
(21 relationships. The above adaptive order M is from the present to the past lN-1) - The number of steps of the approximate measurement signal obtained by expressing the AGC signal Qk from the present time to the time T8 with a step-like function is equivalent to The variance σ1 of the squared interest noise component of the remaining Norse vectors other than the subsequence of the Norse vectors, that is, the higher-order Norse spectrum (selected above)
Output as . (8) is a fast inverse Norse transformer, and the subsequences GO, C1, . . . of the above Norse vector
Perform the inverse Norm transformation of CI1 and move from the present to the past (N-1)
・The optimal approximate measurement signal Fo, which approximates the AGC signal Qk up to time T8 with a step function consisting of M steps;
WB ---, F"-m 1
-v l Wi-2 as average values Yk and Yk-1, respectively
Output as . (9) is the high-speed Haar transformer (6) above.
The adaptive filter A-Tsuru, which is composed of a one-step number determiner (7) and a high-speed inverse Haar transformer (8), is a differential filter that calculates the smoothed value Yk using the smoothed values Yk, Yk-1 and the exponential plane. The differential is calculated and output as a differential signal Wk.

+1) 1 is a discriminator which compares the square value of the differential signal Wk with the variance σ2 output from the staircase determiner (7) and outputs a discriminator signal Dk. The operation of each block in FIG. 2 will be explained in detail below using mathematical formulas and diagrams.

Figure 3 shows the configuration of the DC removal filter. In Figure 1, NA+, MA, ,..., MAI-1 are ND
Converter (storage device f whose memory contents are updated in synchronization with T8 every 21 conversion cycles, cA is a storage device that stores a constant ■,
A D Ds is an adder, and SUB+ is a subtracter.

DI Vl is a divider, and the input corresponding to the denominator is shown by a thick line. Xk&-t A/D converter (2) output, Xt1
, Xk-2, . . . +Xk+x-r are respectively stored in the storage device MAI1MA1. ..., the value stored in MAI-1, ■ is the value stored in the storage device CA.

SAM is the output of the adder ADDt, MEAk is the output of the divider DIV+, and Fk is the output of the subtracter Fk.

In such a configuration, the storage device MA121M
A”-3+..., the contents of MAt are A/D converter (2)
At every conversion period Tl, the adjacent storage device MA1-1,
M Al-2, . . . , MAtK are sequentially transferred in this order, and the digital signal Xk is newly stored in the storage device MA+. signal xk and the storage devices M As , MA7 .・
. . , the values Xk-1, Xk-2°, . . .
is divided by the value stored in the storage device CA in the divider DIVs to produce the signal MEAk. In other words, the storage devices M A+ , M An , ..., MAI-1
) CA.

Adder A D DI , Divider D I Vt Ic Yotte T31 The calculation shown in Equation 1c is performed.

ME Ak = IS AM-” (Xk + Xk
z+...I +X k+1-I)...
......-+31 This is the digital signal Xk, Xk-1,...l) (k+1-) of the past (I-1) 18 hours.
This means that the average value of t was calculated as the signal MF:IAk.

By setting (2) to a sufficiently large value, it operates as an ultra-low-pass filter, and in this case, the signal M E Ak represents the DC component of k (digital signal).Next, the subtractor S
The signal ME is converted from the digital signal Xk by U Bl.
Ak is subtracted (calculation shown in formula 41 is performed).

Fk2Xk −MEAk=Xk −−(Xk + Xk
-1+・・・・・・・・・+Xk+1-71
...... River... (4+This is a digital signal Xk
This indicates that the value obtained by subtracting the DC component contained therein is calculated as the signal Fk. After all, in the DC removal filter configured in this way, the digital signal Xk
The DC component superimposed on is calculated as a signal MEAk by the average value processing shown in equation (3).

Further, as shown in equation (4), the signal MEAk representing the DC component is subtracted from the digital signal Xk to calculate a DC-removed signal Fk from which the DC component has been removed. FIG. 4 shows an automatic gain adjuster (
4) shows the configuration of MBI, MBg in V
, −, MBJ 1 .

and M C+ 9M Ct , ...9M CL 1
is a storage device that updates the memory contents in synchronization with the conversion cycle T8 of the A/D converter (2); CB, CC, and CD are storage devices that store constants J, L, and K, respectively; ADDt, ADD3
5UBv is a subtracter, SD is a sign discriminator that outputs -1 when the input signal is negative and 1 when it is positive, C
a clipper that saturates the LPFi lower limit to the value δ (where δ is a constant), M U Ll , MULt ld , a multiplier;
DIV's, DIV! l, DI
The input corresponding to the denominator of the V41d divider is shown by a thick line. Fk is the output of the DC removal filter (2), Fk-z
, l”k-2.

..., Fk+x-J are the storage devices M Bt, respectively.
, MBt, ..., value stored in MBJ-1 + A Vks , AVlc-2, ---, A V
k+t-L U-tray-'F and the above storage device M C1,
MC, ..., MCL-1) C memorized value,
J, L are the values stored in the storage devices CB, CC, CD, respectively (K is a constant. 1, SX, SA
V are the outputs of adders ADDt, ADD, and MEB, respectively.
k, AGk, Gk each divider DI Vs
, D I Vs , D I V4 O output, AVk, G
k are the outputs of the multipliers M U Lt and M U Lt, respectively, FG is the output of the sign discriminator SD, and CG is the clipper C.
This is the output of LP. In such a configuration, the storage device MBJ-21M BJ-31...

The contents of the MBI are stored in the adjacent storage capacity M BJ-1, M BJ-2, . . . every conversion period T8 of the A/D converter (2).
, M.B.

9 are sequentially transferred in this order, and the DC removal signal pk is newly stored in the storage device MB+.

Similarly, storage devices M CL-2, M CI, -a,...
・The contents of MC+ are also the conversion period T8 of A/D converter (2)
Storage device M CL-1 @ M CL-21 next to each
..., MC5 in this order, and the signal AVk is newly stored in the storage device MCI. Signal 1''k and the storage device MB>, MBt,...-, MBJ-1
-1, Fk-2, ..., pk+ for the value p' stored in
1-J are added by the adder A D Dt to produce the signal SX.
Further, the signal SX is divided by the value J stored in the storage device CB in the divider DIV, and the signal M
It becomes EBk. In other words, storage devices M B+ , M
Bt,..., MBx-1, CB, adder ADD,...
The divider DIVIC performs the calculation shown in equation (5).

M E Bk = "S X =' (Fk + Fk-1
+...J 1i'+x-s) ・・・・・・・・・
...(5) This is past (J-1)/T1) time D
C removal signal Fk, Flcml, -, pk+x-J
This indicates that the average value of is calculated as the signal MEBk.

Next, the above signal MEBk is subtracted from the DC removed signal pk by the subtracter S U Bl to become the signal VR, and the sign of the signal VR is determined by the sign discriminator SD. A signal FG is output, and this and the signal VR are multiplied in a multiplier MULs, and the absolute value of the signal VR is taken, resulting in a signal AVk. Tsumashi subtractor 5UB1. The calculation shown in equation 61 is performed by the multiplier M U Lt and the sign discriminator SD.

Avk= l rk-ME Bk1 = 1Fk bow 1□Fk+1-t... (6) This is the DC removal signal Fk and the past (J-1i ・18 hours of DC
Removal signal Fk, l”k-1,...-, Fk+l-J
This indicates that the absolute value of the difference from the average value of is calculated as the signal AVk.

Similarly, the signal AVk is sent to the storage device MCs, MCx, . . . , MCL by the adder ADDA.
-1 stored values A Vlc-1, A Vk-2, -
, AVk+l-L ,! : The signal SAV is added to the signal SAV, and the signal SAV is further divided by the value stored in the storage device CC in the divider DIv* to form the signal A.
Becomes Gk. Trench storage devices MCI, MC,...
- 9MC'-' + CC, adder ADDA, and divider DIv perform the calculation shown in formula 71.

A Gk--S AV=” I AVk + A Vk
-s +L L...+AVk+x-r,)...
(7) This is the signal AVk of the past (L-1) 18 hours.

AV+1. -, A'Shc+5-L(7) indicates that the average value is calculated as the signal AGk.

Here, AVk shown in equation (6) represents the instantaneous change in the measurement signal X, and further, signal ACk shown in equation (71) represents the instantaneous change in the measurement signal
l! It means a signal averaged over time, and shows the average amplitude of the measurement trace 1 signal X.

R'J yba CLP, divider DI V4, multiplier M
UL, the above D with the signal AGk calculated as above.
C-removed signal Flc divided by L −1 of the above measurement signal
The amplitude of the measurement signal X is adjusted so that the average value over 8 hours becomes constant.

However, the clipper C
In LP, the lower limit value is limited to δ (however, δ is a constant) and the signal GG is not obtained.The DC removed signal Fk is divided by 1r by this signal GG using a divider DI V4 to obtain the DC removed signal Fk. , and is further multiplied by a constant K in a multiplier MUL, to finally calculate an AGC signal Qk adjusted so that the average amplitude of the measurement signal X is constant. This can be expressed as equation (8).

Gk==Up・K −Fk G After all, in the automatic gain adjuster (4) configured in this way, if the amplitude of the measurement signal X, that is, the amplitude of the DC removal signal Fk is large, the signal AGk will also be larger than the formula From equation (8), the above DC removal signal Fk is converted to signal AGk.
The amplitude of the AGC signal Qk divided by (
According to equation 8), the AGC signal is adjusted in the direction of increasing, and the average value of the amplitude of the measurement signal X is adjusted to be constant. The trenched AGC signal Qk is the past (J1L-1) 18-hour DC removed signal pk, Fk as shown in equation (9).
-1, . . . , Fk+t-J-1, are adjusted so that the average value of the amplitude of the measurement signal X is constant.

Gk"'= AGk (Fk, -1 for p',...
Fk+t-y-L"k-(91If there is no DC removal filter (31) in front of the automatic gain adjuster (4), if a DC component is superimposed on the measurement signal The AGC signal GDk corresponding to equation (9) is represented by a button as shown in the open equation, and the DC signal GDk corresponds to the equation (9).
As a result of gain adjustment on the C component, the size changes and becomes a component of waveform distortion, but in this invention, the DC removal filter (3)
Since the DCC components are removed in advance, the waveform distortion as described above can be prevented.

・ (Fk+A) ・・・・・・・・・a(1 (However, A is a constant representing the DC component, FDk=Fk+A
, F Dec-1=Fk-+ + A,...
F Dk+5-J-t, =Fk+1-J-L +
A) Figure 5 shows this situation, and (a) and (b) in the figure show a DC removal filter (31) in front of the automatic gain adjuster (4).
The figure shows the AGC signal Qk when the DC component A is not superimposed on the digital signal Xk and when the DC component A is superimposed on the digital signal Xk, respectively.

In the end, the DC removal filter (3) and the automatic gain adjuster (4)
) so that the average value of the amplitude of the measurement signal The above measurement signal
can be smoothed.

Next, the adaptive filter (9) will be explained. Note that FIG. 6 is a diagram for explaining the principle of approximation and smoothing of a calculated signal using a stepwise function.

First, the memory value Fo stored in the data buffer (5),
Wl,...-,V'N-:x-...tiD(Tatal, Fo:r Gk-N+1, vI =Ql
c-N+2,-.

ViN-2-=Qk-1, V'N-1=Qlc) to M
A step-like function consisting of steps is fitted using the least squares method, and the sequence (sequence indicating the height of the stairs) is written as Wo, Fl, ... -, FM-1 ...
...az, then the approximation error, that is, the error between the numerical sequences of the +1))+12 formula can be expressed as the 0Hu) formula.

However, N = 2'', M = 2m, and n is a positive integer)
Let it be m.

In this case, mathematically, the inequality of the equation 9 holds clearly.

PO≧P1≧・・・・・・≧Pm ・・・・・・−・
...(US) If we give a positive number Σ as the allowable value, Pm≦
Since there is a smallest m in Σ, this is used as an index.

−Tomi M=2 ・・・・・・・・・(I
If Q is the sequence Wo, Wl,..., Fi-1...
...tiη is a step function with the error Pm less than or equal to Σ and has the smallest number of steps, and is the optimal approximate measurement signal Fo, Wl that optimally approximates the value of the AGC signal Gk from the present to the past (N-1>・TII) ,...-, FM-1 are obtained.This optimal approximate measurement signal FO, Fl,...

The current value of i-1 - 1 and the previous value F
Each time the AGC signal Qk, which is stored in the data buffer (5) with M-s as smoothed values Yk and Yk-1, is updated every cycle Ts, the optimal approximation measurement (i %
'O, Ft, -, FM-1f calculated, smoothed value Yk+
If Yk 1 is updated, the smoothed value Yk, Yk-1l'i
When the adaptive order M is small, the output is obtained by greatly smoothing the changes in the AGC signal Gk, but as M becomes larger, the degree of smoothing becomes smaller. When M=N, the smoothed values Yk and Yk-t become the AGC signals Gk and Gk-1, respectively. Become that thing. That is, the output Fo of the data buffer (5), γ1. ...
, when there is no pulse-like signal (1) indicating an abnormality in FN-1, the adaptive order M becomes small and the noise is greatly smoothed, but
When the pulsed signal (1) is included, the adaptive order M becomes large, the smoothing effect becomes small, and the pulsed signal fi+ is output as is. The trenched AGC signal Qk is passed through a type of adaptive filter whose passband changes depending on its characteristics, thereby achieving effective filtering with little delay.

As explained above, the adaptive filter that approximates the stored values Fo, Fl, ..., FN-1 stored in the data buffer (5) is a fast Haar transformer (6), a step number determiner (
7) and a high-speed inverse Haar transformer (8). First, the high-speed Haar converter 16+ is a data buffer (5)
The stored values 10, Wl, .

Outputs CN-1. Here, the Haar function x is the interval ro,
t] is a function determined as follows. Let t be a variable in the interval [0, 1) from 1 to xn(t) = 1 0
≦t≦1 ・・・・・・0 rise・・・・−・・・・・・
...α9 Figure 7 depicts a part of the Haar function of (+16 equation (1)). Using this Haar function, the above AGC signal Wo+ Ws+ −+ FN-1aG!Q equation Te table M, f
*It is also known as Ruco.

・・−・−・・・・・・・・・・・・■ (j = 0.1.=-, N-1:N=2”) where co
, Czld is the Haar spectrum, and its relationship with the Haar spectrum Co, Ca, ..., CN-1 is Ca1
) is as follows.

c2n-□=cn(2n-1) (1=1.2.-=, n: k=1.2.--,
2L”) Furthermore, using the formula ■, the above memorized values Fo, F
l...

The following fast Haar transform is used to obtain the Haar spectra Co, CL from FN-t. First, the above memory value W
o, Wl,...-, FN-1 are regarded as a sequence and the following recurrence formula is calculated.

io+Fj-Fj, Ij=0.1.
-, N-1) -H fin - W button゛ゝ+Fj print:) (p-), 2. ..., nl:q==Q,t,...,
2-1)・・・・・・・・・・・・(c) Then obtained. , [0), , using fn-tl
ln-1) 2 Co == Fo -1-Fl
・・・・・・・・・・・・(Foundation) −2・2・−
・. n-! ::)=Fl:””)-F:))810
09.0S (r=1.2.-, n; S=Q, 1.-.2
-1) From the throat formula, the sequence c, (2'Ll), , , ,
By sequentially calculating c, , co, the Haar spectrum of the QTJ equation, that is, Co, CI, ・=, C
n-1 is found.

In this way, the high-speed Haar transformer (6) has the function of performing calculations of equations (1) to (3). It is also known that mathematically. Therefore, if the Haar spectrum Ct is found, it is easy to find the index i that is the minimum m that is punishable by Pm≦ε, where the error/'m is a predetermined tolerance value ε. It is possible.

The step number determiner (7) calculates the index ball below the tolerance value 8 from equations (A) and (5), outputs the error/'m at this time as the variance σ3, and uses the equation (2) from the index gate. The order M shown in is obtained, and the subsequences Co, CI, . . . , cy-z of the Haar spectrum are selected and output based on the order M.

Next, the high-speed inverse Haar transformer (8) performs an inverse Haar transform on the subsequences CO, CI, .
, F) Calculate J-1. Transforming Kazuzu Matsu(c) formula, Fofm-1)=yo(CO+C'lゝ)・-・・・・・・・
... (to) M q, (m-') = (0 (101' to 2M...) is obtained. Sequentially) 61) If you calculate the formula % formula %! , = ψ = ψ (j=o, t,..., M-
1) IJ ... (to) multiple sequences Vy, , F1. ......'Q-+ is obtained. The high-speed inverse Haar transformer (8) executes calculations from equations (2) to (2), and outputs FM-1, -12, respectively.

Next, the differential filter αq is the high-speed inverse Haar transformer (8)
The differential signal ωk is calculated from the smoothed value Yk+Yk−x, which is the output of , and the exponential system, which is the output of the step number determiner (7).

wk=(yk-Yk-+)/2 ・Ts) ・・
(To) However, n is a value that determines the number of 1-ball spectra (expressed by the following formula), and Ts is a sampling period for sampling the measurement signal X.

The discriminator (b) uses the differential signal W which is the output of the differential filter.
A pulse-like signal (1) indicating an abnormality included in the measurement signal X using k and the variance σ2 which is the output of the step number determiner (7)
Detect.

Memory value ψ0 stored in data buffer (5). ψ.

, status, ..., a hull signal indicating an abnormality in car 1 (
1) If there is no noise and there is only noise, the above smoothed values Yk, Yk-1
is a low-frequency smooth signal, so the differential signal Wk obtained from Yk and Yk-1 is small and if its square value is Uk, then Uk falls within the range of α times the dispersion σ2. If this is expressed as a formula, it will look like the following formula. However, α is a constant determined from experimental results.

Uk〈ασ2...[Yes] Conversely, the stored value ψ stored in the data buffer (5). ,ψ1
．． ......If there is a pulse-like signal (1) indicating an abnormality in ψN-1, -1 is output as a steep pulse-like signal to the smoothed value Yk"', so its differential signal is Since Wk is also large, its square value Uk is also large.This can be expressed as the following equation.

Uk≧ασ2 ... (to) The above memorized value ψ0 is maintained while the trowel formula is established. ψ□,...
. . . Assuming that there is no pulse-like signal (1) indicating an abnormality in ψN-1, the discrimination signal is set to = 0, and conversely, when the formula (to) holds, it is notified as = 1, and the above stored value ψ0゜ψ8 is set. ．．
...Pulse signal indicating abnormality in ψN-1 (1)
It can be detected that there is. Moreover, the above memory value ψ0
．． ψ1. ..., ψN-1 is the D of the measurement signal
AGk C-double type adjusted so that the average value of the amplitude is constant without causing waveform distortion due to the C component; Since it is memorized, the detection result does not depend on the fluctuations in the amplitude of the measurement signal X and the DC component as in the past. You will no longer be influenced by it.

In this way, the discriminator CL,) uses the square value Uk of the differential signal and the variance σ2 to determine the equation) at t34, and calculates the measurement signal X.
A determination signal Dk indicating whether or not a pulse-like signal indicating an abnormality is present is output. FIG. 8 shows this situation.
AGO signal Gk, smoothed value Y2. For the differential signal, the square value U of the differential signal.

, the distribution σ2 and the approximate shape of the detection signal Dk.

As described above, according to the present invention, the Do acid component contained in the vibration acceleration type measurement signal is removed by a removal filter in advance and then inputted to the automatic cane adjuster, thereby preventing waveform distortion caused by the DC component. Next, the amplitude of the measurement signal is adjusted to a constant value, the high frequency noise contained therein is removed by an adaptive filter configured by combining Haar transform and inverse Haar transform, and the steepness of the waveform is removed by a differential filter. Since the presence or absence of an abnormal point is determined by obtaining a differential signal representing the temperature, the abnormal point can be detected without being affected by fluctuations in the amplitude of the measurement signal and the Do acid component.

Here, as an example, an example is shown in which a vibration accelerometer is used to detect an abnormality in a power generation device.

The present invention is not limited to this, and can be applied to abnormality detection in any equipment, device, or natural phenomenon.

〔Effect of the invention〕

This invention uses the above-described conventional abnormality detection device consisting of the automatic gain adjuster, the adaptive type failure, and the differential failure to calculate the average value of the measurement signal over a certain period of time from the measurement signal. DC to remove Do acid component by pulling
By adding a removal filter to remove in advance the DC component superimposed on the measurement signal, the automatic gain adjuster can prevent waveform distortion caused by the DC component and detect abnormal points by eliminating fluctuations in the amplitude of the measurement signal. This has the effect of further increasing the ability to detect abnormal points without being influenced by DC components.

ζ Brief explanation of the diagrams Figure 1 is a diagram showing measurement signals from a vibration accelerometer installed in a power generation device, Figure 2 is a diagram showing an embodiment of the present invention, and Figure 3 is a diagram showing a DC removal filter (3 ), FIG. 4 is a diagram showing the configuration of the automatic gain adjuster (4), and FIG. 5 is a diagram showing the configuration of the automatic gain adjuster (4).
Figure 6 is a diagram showing the DO removal effect of the o-removal filter, Figure 6 is a diagram showing the principle of approximation and smoothing of the measurement signal by a step function, and Figure 7 is
The figure shows a part of the Haar function, and FIG. 8 is a diagram showing the shapes of the DC removed signal, AGO signal, smoothed value, differential signal, square value 1 variance of the differential signal, detection signal, etc. for the measurement signal.

In the figure, +1) is a pulse signal indicating an abnormality, (2) is a ~Φ converter, (3) is a Do removal filter, (4) is an automatic gain adjuster, (5) is a data buffer, and (6) is a high speed Haar transformer, (7) is step number determiner, (8) is expressway no.
(9) is an adaptive filter, (1) is a differential filter, (b) is a discriminator, MA□, MA,...
...' MAI-1, MBl, MBz, ...
・, MBJ-1, MO□+ M O! ＋・・・・・・
・,MOL-,,OA,OB,00. OD is a storage device.

A D D 1+ A D Dt, A D D 3 is an adder, 5UB1.5UBt is a subtraction N+ DIvl
r DIVs + DIVs r DIV4 is a divider,
SD is a sign discriminator, OLP is a clipper.

MUL, , MULx are multipliers. In addition, the same or equivalent parts in the figures are indicated with the same reference numerals.

Claims (3)

[Claims] (1) A DC removal filter that takes measurement signals obtained from various sensors as input and removes DC components by subtracting the average value of the measurement signals over a certain period of time from the measurement signals; and the DC removal filter removes the DC components. an automatic gain adjuster that adjusts the amplitude of the measurement signal based on the value obtained by adding and averaging the measured signals over a certain period of time in the past; and updating the measurement signal whose amplitude has been adjusted by the automatic gain adjuster at regular intervals. a data buffer to be stored while
r) Error between a high-speed Haar transformer that converts and calculates a plurality of Haar spectra, and the measurement signal stored in the data buffer and the stepped approximate measurement signal obtained from the inverse Haar transformation of the subsequence of the Haar spectrum. is calculated as the variance, and the subsequence of the Haar spectrum whose error is below the set tolerance value and closest to the tolerance value is selected, and the step number determiner calculates its order, and the subsequence of the Haar spectrum selected above is Adaptation that consists of a high-speed inverse Haar transformer that calculates the optimal approximate measurement signal by inverse Haar transform, and calculates the optimal approximate measurement signal each time the stored value of the data buffer is updated, and outputs it as a smoothed value. an abnormality detection device comprising: a differential filter that calculates a differential signal of the smoothed value by inputting the smoothed value and the order; and a discriminator that compares the square of the differential signal with the variance. (2) In place of the differential filter, a square circuit is provided which takes the smoothed value as input and squares the smoothed value, and compares the squared value of the smoothed value with the variance to detect abnormalities that suddenly appear in the measured signal. An abnormality detection device according to claim (1), which detects a peculiar signal. (3) the automatic gain adjuster is configured to adjust the amplitude of the measurement signal based on a standard deviation over a certain period of time in the past;
An abnormality detection device according to claim (1), which detects an abnormality or a unique signal that suddenly appears in the measurement signal.