JP3387331B2 - Abnormal signal detector and abnormal signal source detector - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a filter characteristic and a filter cut component obtained by performing adaptive digital filter processing on a one-dimensional time-series signal such as sound or vibration obtained in, for example, plant equipment. The present invention relates to a device for detecting an abnormal signal and a device for detecting an abnormal signal source using the same.

[0002]

2. Description of the Related Art Conventionally, in this type of abnormal signal detecting apparatus, first, a signal obtained in a normal state is subjected to Fourier analysis and the result is stored as a database. Then, the normality / abnormality at the time of inspection is detected by comparing the result of Fourier analysis of the signal obtained at the time of inspection with the data stored in the database. Here, the signal obtained under normal conditions is a stationary signal.

Next, Fourier analysis will be described. In Fourier analysis, generally, a time series signal obtained by A / D converting an analog signal into a digital signal is subjected to discrete Fourier transform to obtain a spectral density distribution. Usually, this spectral density distribution is averaged several times. Discrete Fourier transform is described in, for example, Hiroshi Miyagawa, Kenichi Kido et al., "Digital Signal Processing", 15th edition, pp. 10 to 20. When an input signal at an arbitrary time t is x (t), a Fourier transform X obtained by converting x (t), which is a time function, into a frequency function
(f) is expressed by Equation 1. Where f is the frequency.

[0004]

[Equation 1]

The signal actually handled is a finite discrete signal which has been A / D converted. X _p, which is a discrete signal of the input signal x (t)
Discrete Fourier transform X _{k of} (p = 0, ..., N−1)
(K = 0, ..., N−1) is represented by Expression 2.

[0006]

[Equation 2]

Here, p and k are integers, and N is the number of discrete signals x _p . The spectral density distribution S _k is _obtained from the discrete Fourier transformed X _k by Equation 3.

[0008]

[Equation 3]

In Expression 3, Re (X _k ) is a function representing the real part, and Im (X _k ) is a function representing the imaginary part.

Next, an example of anomaly detection by Fourier analysis will be given by taking sound in a plant as an example. FIG. 29 shows a normal signal of sound at a normal time in a plant.
It is a waveform sampled by KSPS. In the figure,
The vertical axis is amplitude and its unit is Pa, and the horizontal axis is time and its unit is second. FIG. 30 shows the normal signal shown in FIG. 29 for every 256 points.
It is the average of the spectral density distributions obtained by performing Fourier transform 00 times. In FIG. 30, the vertical axis represents the spectral density distribution, the horizontal axis represents frequency, and the unit is Hz. Further, FIG. 31 shows a waveform obtained by sampling an abnormal signal obtained by adding an abnormal sound to a normal sound in this plant at 48 KSPS. In FIG. 31, the vertical axis represents amplitude and the unit is Pa, and the horizontal axis represents time and the unit is second. FIG. 32 is an average of spectral density distributions obtained by Fourier transforming the abnormal signal shown in FIG. 31 every 256 points 100 times. In FIG. 32, the vertical axis represents the spectral density distribution, the horizontal axis represents frequency, and the unit is Hz. Further, FIG. 33 shows the spectral density of the abnormal sound added in FIG. 31, which is obtained by Fourier transforming every 256 points 100 times. The vertical axis represents the spectral density distribution, the horizontal axis represents frequency, and the unit is Hz. Since the scale of the spectral density distribution in this figure is different from those in FIGS. 30 and 32, the vertical axis does not include numerical values.

FIG. 3 is a spectral density distribution of an abnormal signal.
When 2 is compared with FIG. 30, which is the spectral density distribution of the normal signal, a very small abnormal signal component is seen around 15 KHz, and the power is slightly increased around 10 KHz. This difference matches the spectral density distribution of the abnormal sound in FIG. The spectral density in FIG. 33 is before the generation of the abnormal signal, and the measuring system is different from that at the time of measuring the abnormal signal, and therefore the scale is different from that in FIGS. 30 and 32.

In the abnormality detection by Fourier analysis, FIG.
The absolute value of the difference between the spectral density distributions of the normal signal shown in Fig. 32 and the abnormal signal shown in Fig. 32 is set as a threshold value which is twice or three times the standard deviation when the spectral density distribution is measured a plurality of times. FIG. 34 is an absolute value of the difference between each frequency component of the spectral density distribution of the normal signal shown in FIG. 30 and the spectral density distribution of the abnormal signal shown in FIG. 32 (solid line 361).
31 is a graph of twice the standard deviation (broken line 362) of the spectral density distribution of the normal signal of FIG. 30 measured multiple times. In FIG. 34, the vertical axis represents the spectral density, the horizontal axis represents the frequency, and the unit is Hz.

As is apparent from FIG. 34, the absolute value of the difference between the spectral densities of the normal signal and the abnormal signal (solid line 361) is a small value, which is twice the standard deviation which is the threshold value for detecting the abnormality ( Comparing with the broken line 362), it is only possible to detect or not to detect the abnormality. Especially when the number is tripled, no abnormality can be detected. Anomalies can be detected if the threshold value is made smaller, but conversely false alarms are likely to occur during normal operation.
In addition, since the Fourier analysis is an analysis for a stationary component, it cannot handle sudden abnormal signals.

Next, the adaptive digital filter processing which is the basis of the present invention will be described. FIG. 35 is a basic configuration diagram of the adaptive digital filter, in which 161 is an adaptive digital filter processing unit and 162 is an adder. The operation of FIG. 35 will be described below. In FIG. 35, the input signal x _k is subjected to convolution processing in the adaptive digital filter processing section 161 and becomes an output y _k . And adder 162
Adding the inverse and the target signal d _k, the sign of the output y _k by the error ε as the difference between the target signal d _k and the output y _k
get _k . After that, the error ε _k is passed to the adaptive digital filter processing unit 161. By repeating the above processing recursively, a filter that minimizes the error ε _k is generated in the digital filter processing unit 161.

Next, the adaptive digital filter processing unit 161 will be described. The filter coefficient sequence of k sample time is W
_{Let k} . Hereinafter, the filter coefficient string is simply referred to as a filter coefficient. The filter coefficient is obtained by recursively repeating Expression 5 using the gradient ∇ _k shown in Expression 4 of the evaluation function D (W _k ) of the system. In Expression 5, μ is a parameter that controls the magnitude of the correction amount in each repetition, and is called a step parameter.

[0016]

[Equation 4]

An algorithm using the instantaneous squared error ε _k ² itself as an evaluation function is an LMS (Least Mean).
This is called the Square algorithm. The estimated amount of the gradient ∇ _k by the instantaneous squared error ε _k ² is expressed by Equation 6.

[0018]

[Equation 5]

Here, X _k is an input signal sequence at k sample times. Further, ε _k is expressed by Equation 7, and W of Equation 7
_k ^T X _k represents the convolution of the filter coefficient and the input signal. Using an estimator of the gradient ∇ _k according to equation 6, an LMS algorithm for determining the filter coefficient is represented by equation 8.

[0020]

[Equation 6]

Time-varying estimation value σ of the average power of the input signal
Introducing _k ² and new parameters u and α, Equation 5 is expressed as Equation 9.

[0022]

[Equation 7]

Where L is the order of the filter, L + 1
Is the filter length. By setting α ≠ 0, the time-varying estimated value σ _k ² of the average power of the input signal can be adapted even when the input signal power changes. This adaptive digital filter processing is usually used for noise removal, plant model estimation, and the like.

Next, the signal source detection method which is the basis of the present invention will be described. A method for detecting the direction of the signal source is described, for example, in "Acoustic System and Digital Processing" by Oga, Yamazaki, and Kaneda, first edition, pages 197 to 199. FIG. 36 is a diagram for explaining a method of calculating the direction of a signal source from two input sensors on a two-dimensional plane.
1a and 291b are microphones, 292 is two microphones 29
The distance difference between 1a and 291b, 293 is the arrival direction θ _s of the sound wave.
In this example, it is assumed that sound is set as the signal source, the input sensor is a microphone, and the sound of the signal source is transmitted in the air.

A method for detecting the direction of the signal source on the two-dimensional plane from the signals obtained by the two input sensors 291a and 291b will be described below. In FIG. 36, let us consider that a plane wave arriving from the θ _s direction is received by two microphones installed at a distance d. This time,
Between the received sound signal x ₁ (t) and x ₂ (t) of each microphone,
The relationship of Expression 10 is established. Where c is the speed of sound and d is the distance
It is 292. Therefore, conversely, if the time difference τ _s between the sound reception signals x ₁ (t) and x ₂ (t) is known, the arrival direction θ _{s of the} sound wave can be calculated by Equation 11
Can be obtained by

[0026]

[Equation 8]

Next, the time difference τ _s is obtained. The time difference τ _s can be obtained from the cross-correlation function φ ₁₂ (τ) between the received signals x ₁ (t) and x ₂ (t). Substituting equation 10 into equation 12 defining the cross-correlation function yields equation 13.

[0028]

[Equation 9]

However, E [•] represents an expected value, and φ ₁₁ (τ)
Represents the autocorrelation function of x ₁ (t). Autocorrelation function φ
It is known that ₁₁ (τ) takes the maximum value when τ = 0. Therefore, φ ₁₂ (τ) takes the maximum value at τ = τ _s . From this, if the cross-correlation function φ ₁₂ (τ) between x ₁ (t) and x ₂ (t) is calculated and τ that gives the maximum value is obtained, τ _s is obtained, and this τ _s is expressed by Substituting into 11, the arrival direction of the sound wave θ _s
Is required.

FIG. 37 is a diagram for explaining a method of calculating the position of a signal source from three input sensors on a two-dimensional plane, where 291c is a microphone and 301 is a signal source. In this way, three input sensors 291a, which are not on a straight line,
A method of detecting the position of the signal source on the two-dimensional plane from the signals obtained by 291b and 291c will be described. Figure 3
In 7, the time the sound from the signal source to the microphone 291a reaches a t _a. Similarly, it is assumed that the time at which the sound reaches the microphone 291b is t _b and the time at which the sound reaches the microphone 291c is t _c . In addition, τ _ba is the microphone 29
Assuming that the sound arrival time difference between 1b and the microphone 291a, τ _ca is the sound arrival time difference between the microphone 291c and the microphone 291a, and the squares of the distances are equal, the simultaneous equations of Expression 14 can be obtained.

[0031]

[Equation 10]

Here, the time differences between the microphones τ _ba and τ
_ca can be obtained by the cross-correlation function shown in Expression 13. The simultaneous equations of Equation 14 have three unknowns, x, y, and t. Since there are three equations, a solution can be obtained, and the signal source (x, y) can be detected.

FIG. 38 is a diagram for explaining a method of calculating the position of a signal source from four input sensors in a three-dimensional space, and 291d is a microphone. in this way,
4 input sensors 291a, 291b, 291c, which are not on the same plane
A method of detecting the position of the signal source in the three-dimensional space from the signal obtained by 291d will be described. In FIG. 38, the time taken for the sound to reach the microphone 291a from the signal source is t _a . Similarly, the time for sound to reach the microphone 291b is t _b , the time for sound to reach the microphone 291c is t _c , and the time for sound to reach the microphone 291d is t.
_{Let d} . Furthermore, τ _ba is the time difference of arrival of sound between the microphone 291b and the microphone 291a, and τ _ca is the microphone.
If the sound arrival time difference between the microphone 291c and the microphone 291a and τ _da is the sound arrival time difference between the microphone 291d and the microphone 291a, the simultaneous equations of Expression 15 can be obtained.

[0034]

[Equation 11]

The simultaneous equations represented by the equation 15 have four unknowns, x, y, z, and t, and since there are four equations, the solution can be obtained, and the signal source (x, y, z) can be obtained. Can be detected.

Next, a method of obtaining a solution from the equation 15 will be described. Since the simultaneous equations of Equation 15 are difficult to solve analytically, the solution is obtained by numerical analysis such as Newton-Raphson method. The Newton-Raphson method is described, for example, in Akasaka, "Numerical Calculation," 22nd edition, pages 223 to 225. First, a case where a simultaneous equation with three unknowns is obtained by the Newton-Raphson method will be described.
In general, the equation including three unknowns is set as Expression 16.

[0037]

[Equation 12]

Equation 17 is the nth-order approximation of Equation 16 (x _n , y _n , t
It is a Taylor expansion near _n ).

[0039]

[Equation 13]

Where (x−x _n ), (y−y _n ),
By omitting the second-order or higher-order terms of (t−t _n ), and setting it as Expression 18, Expression 17 becomes Expression 19.

[0041]

[Equation 14]

[0042]

[Equation 15]

The correction amounts Δx _n , Δy _n , and Δt _n can be obtained by solving the simultaneous linear equations of the equation (19). The obtained correction amount is substituted into the equation 18 to obtain x _{n + 1} , y _{n + 1} and t _{n + 1} , and further substituted into the equation 19 to obtain the next correction amount. Formula 18
By giving the initial values x ₀ , y ₀ , and t ₀ to Equation 19 and recursively repeating until the correction amount becomes smaller than a predetermined value, an approximate solution is obtained. The solution to be obtained is x _n , y _n , and t _n at the time when the recursive process is completed. When there are four unknowns, the equation is the same as when there are three unknowns described above, and the equation includes four unknowns.

When the signal source is detected by the above-mentioned signal source detecting method, it cannot be obtained unless the arrival time difference of the signals detected by the plurality of input sensors is known. However, since the spectrum density distribution is obtained from the input signal in the abnormality detection method by Fourier analysis, the input signal does not remain in the form of a time series signal. For this reason, the arrival time difference at the plurality of input sensors cannot be obtained, and the direction and position of the signal source of the abnormal signal cannot be obtained.

[0045]

As described above, according to the method by Fourier analysis, it is difficult to set the threshold value when the abnormal component is small. Therefore, it is difficult to detect the abnormality such that false alarm is likely to occur when the abnormality is detected. There was a problem. Also,
There was a problem that it could not handle sudden abnormal signals. Further, even if the abnormality detection method based on the Fourier analysis can detect that an abnormality has occurred, it has not been possible to obtain the direction and position of the abnormality signal.

The present invention has been made to solve the above problems, and an object of the present invention is to obtain an abnormal signal detecting device capable of detecting an abnormal signal even when the abnormal component is small or a sudden signal. Another object is to obtain an abnormal signal source detection device capable of detecting the direction and position of the abnormal signal source in which the abnormal signal is generated when the abnormal signal is detected.

[0047]

An abnormal signal detecting apparatus according to the first aspect of the present invention includes an adaptive digital filter for simulating the characteristics of a delay device with respect to an input signal which is a one-dimensional time series signal, and has a normal function. An abnormal signal for detecting an abnormal signal by comparing the filter characteristic in the process of forming the filter characteristic of the adaptive digital filter with respect to the input signal at the time with the filter characteristic in the process of forming the filter characteristic of the adaptive digital filter with respect to the input signal at the time of abnormality It is provided with a detecting means.

In the abnormal signal detecting apparatus according to the second configuration of the present invention, in the first configuration, it is determined whether the filter characteristic in the filter characteristic forming process of the adaptive digital filter with respect to the normal input signal is appropriate. A normal-time filter characteristic determining unit for determining the normal-time filter characteristic determining unit is configured to use the filter characteristic determined to be appropriate by the normal-time filter characteristic determining unit as the normal-time filter characteristic to be compared by the abnormality detecting unit. is there.

In the abnormal signal detecting apparatus according to the third aspect of the present invention, in the configuration according to the first or second aspect, the filter characteristic in the process of forming the filter characteristic of the adaptive digital filter with respect to the normal input signal is stored. The filter characteristic storing means is provided, and the filter characteristic stored in the filter characteristic storing means is used as the normal filter characteristic to be compared by the abnormality detecting means.

Further, the abnormal signal detecting apparatus according to the fourth configuration of the present invention is, in the configuration according to the first or second aspect, a filter characteristic storage for storing filter characteristics for a plurality of normal input signals according to circumstances. As a filter characteristic in a normal time to be compared by the abnormality detecting means, a filter characteristic in a situation close to the situation at the time of abnormality detection is used among a plurality of filter characteristics stored in the filter characteristic storing means. It is a thing.

The abnormal signal detecting apparatus according to the fifth aspect of the present invention includes an adaptive digital filter for simulating the characteristics of a delay device for an input signal which is a one-dimensional time series signal,
A filter coefficient storage means for storing a filter coefficient in the filter characteristic forming process of the adaptive digital filter with respect to a normal input signal, and a normal component is cut from the input signal at the time of abnormality detection by using the normal filter coefficient. It is provided with a filter cutting means for obtaining a filter cutting component and an abnormality detecting means for detecting an abnormal signal based on the filter cutting component which is an output signal of the filter cutting means.

The abnormal signal detecting apparatus according to the sixth aspect of the present invention is the abnormal signal detecting apparatus according to the fifth aspect, further comprising filter coefficient storing means for storing filter coefficients for a plurality of normal-time input signals according to circumstances. As a normal-time filter coefficient used when cutting by the filter cutting means, a filter coefficient in a situation close to the situation at the time of abnormality detection is used among a plurality of filter coefficients stored in the filter characteristic storage means. is there.

Further, according to the seventh aspect of the present invention, there is provided the abnormality signal detecting device, wherein when the abnormality detecting means of any of the first to sixth configurations detects an abnormality a plurality of times in succession. Is configured to output.

Further, the abnormal signal source detecting device according to the eighth configuration of the present invention is the fifth configuration, further stores a plurality of sensors and a plurality of filter cut components in the same time zone based on the plurality of sensors. Means for calculating the position of the abnormal signal source by calculating the time difference between the filter cut components based on the plurality of sensors when an abnormal signal is detected.

[0055]

DETAILED DESCRIPTION OF THE INVENTION

Embodiment 1. Hereinafter, the abnormal signal detection method by the adaptive digital filter processing according to the first embodiment of the present invention will be described. FIG. 1 is a configuration diagram in the case of generating a filter that passes an input signal with an adaptive digital filter,
161 is an adaptive digital filter processing unit, 162 is an adder,
171 is a delay device. As shown in the figure, the delay unit 171 delays the input signal x _k by a fixed sample time to obtain the target signal d _k . Normally, the delay device 171 delays by a time corresponding to about 1/2 of the filter coefficient length. On the other hand, the signal x _k input to the adaptive digital filter processing unit 161 is convoluted in the adaptive digital filter processing unit 161 to become the output y _k , and the adder 162 inverts the sign of the target signal d _k and the output y _k. Is added,
The error ε _k is obtained as the difference between the target signal d _k and the output y _k . After that, the error ε _k is calculated by the adaptive digital filter processing unit 16
Passed to 1. By repeating the above processing recursively, a filter that minimizes the error ε _k is generated in the digital filter processing unit 161.

FIG. 2 shows an example of filter characteristics in the process of generating the filter for passing the normal signal shown in FIG. 29 by adaptive digital filter processing. In Fig. 2, the vertical axis is the spectral density, the horizontal axis is the frequency, and the unit is Hz.
Is. In the figure, the solid line 371 is the filter characteristic after 0.1 second adaptive digital filter processing, the short wave line 372 is the filter characteristic after 1 second adaptive digital filter processing, and the long dashed line 373.
Indicates the filter characteristic after the adaptive digital filter processing for 5 seconds, and the filter characteristic after the adaptive digital filter processing for 0.1 second, 1 second, and 5 seconds is also shown.

The filter characteristic indicates the ratio of passing each frequency component, and has a value of 0 to 1 although there is some overshoot. The frequency component whose value is 1 is completely passed,
When the value is 0, it does not pass at all. If the value is 1 or more, amplification will occur. Looking at the process in which the filter is generated in FIG. 2, it can be seen that the filter is sequentially generated from the frequency including a large component in the spectral density distribution of the normal signal in FIG. When the adaptive digital filter processing unit 161 shown in FIG. 1 performs the adaptive digital filter processing to pass the input signal, the generated filter characteristic simulates the characteristic of the delay device 171, and finally the all-pass delay is obtained. Generate a filter. If an all-pass delay filter is generated, the response of all the signals will be equivalent, and the function of the filter will be lost. Therefore, in this embodiment, the filter characteristic in the process of generating the filter is used.

Although the delay device 171 has a pure delay characteristic here, it may have another characteristic as long as it has a characteristic that can be simulated by the adaptive digital filter processing section 161. Any type of delay device with finite length response (FIR) can be used. Also, to use the filter characteristics in the process of filter generation as a database, there is some universality in the filter characteristics after the same time adaptive digital filter processing for normal signals measured at different times. Good.

FIG. 3 shows three filter characteristics which are obtained by performing adaptive digital filter processing for 2 seconds on normal signals measured at different times. In FIG.
The vertical axis represents the spectral density, the horizontal axis represents the frequency, and the unit is Hz. Also, the solid line 381a is the filter characteristic after the normal signal a is adaptive digital filtered for 2 seconds, the short wave line 381b is the filter characteristic after the normal signal b is adaptive for 2 seconds, and the long dashed line 381c is the normal signal c is adaptive for 2 seconds. It is a filter characteristic after digital filter processing. The normal signal a 1, the normal signal b, and the normal signal c are normal signals measured at different times, and the normal signal a 1 is the normal signal shown in FIG.

It can be seen from FIG. 3 that the filter characteristics after the same time adaptive digital filter processing are almost equal, that is, there is some universality. Although FIG. 3 shows only the filter characteristic after the adaptive digital filter processing for 2 seconds, the same result is obtained even if the processing time is not 2 seconds and the same processing time elapses.

Next, the difference in filter characteristics between the normal signal shown in FIG. 29 and the abnormal signal shown in FIG. FIG. 4 shows two signals for the normal signal shown in FIG. 29 and the abnormal signal shown in FIG.
It is a superposition of filter characteristics after adaptive digital filter processing for a second. In FIG. 4, the vertical axis represents the spectral density, the horizontal axis represents the frequency, and the unit is Hz. Also, the solid line 391
Is the filter characteristic after the abnormal signal is adaptively digitally filtered for 2 seconds, and the broken line 381a is the filter characteristic after the normal signal is adaptively digitally filtered for 2 seconds. It can be seen that there is a difference in the filter characteristics between the normal signal and the abnormal signal.

FIG. 5 shows the absolute value of the difference between the filter characteristics of the normal signal and the abnormal signal in FIG. 4 after the adaptive digital filter processing for 2 seconds. In FIG. 5, the vertical axis represents the spectral density, the horizontal axis represents the frequency, and the unit is Hz. Further, FIG. 6 is obtained by superimposing the three absolute values of the difference in the filter characteristics after the adaptive digital filter processing for two seconds on the three normal signals shown in FIG. In FIG. 6, the vertical axis represents the spectral density, the horizontal axis represents the frequency, and the unit is Hz.
Further, a solid line 411 is an absolute value of a difference in filter characteristics between the normal signal a and the normal signal b, a short wave line 412 is an absolute value of a difference in filter characteristics between the normal signal a and the normal signal c, and a long broken line 413.
Is the absolute value of the difference in filter characteristics between the normal signal b and the normal signal c.

Comparing FIG. 5 and FIG. 6, the absolute value of the difference between the normal signal and the abnormal signal shown in FIG. 5 is significantly larger than the absolute value of the difference between the normal signals shown in FIG. It can be seen that it is. It was difficult to detect anomalies for small abnormal signal components by the Fourier analysis method, but the above-mentioned method by adaptive digital filtering has a significant difference between the filter characteristics of the normal signal and the filter characteristics of the abnormal signal. Therefore, it is easy to detect. This is because the adaptive digital filtering process of FIG. 1 produces a filter that allows the frequency component to pass even if the input signal component is a small component, and a filter having different characteristics for different input signals. Is generated. The characteristics of the filter generated by the adaptive digital filter processing are Fourier transformed to obtain the spectral density distribution of the filter coefficient.

In the present embodiment, an abnormal signal is detected by utilizing the filter characteristic in the process of forming the adaptive digital filter as described above, and when the all-pass filter is finally formed, the abnormal signal is detected. Although the detection ability is lost, the processing time of about 2 seconds was appropriate for the signal in the band of about 10 KHz used in this embodiment. However, this processing time is not limited to a narrow range, and the abnormality can be effectively detected as long as it is within the time for forming the all-pass filter.

In order to detect an abnormality by the above adaptive digital filter processing, the average value of the filter characteristics of the normal signal for several times is stored in the database as the filter characteristic of the normal time, and the filter of the normal signal is checked at the time of inspection. It suffices to obtain the filter characteristic after performing the adaptive digital filter processing for the same time as when the characteristic is obtained, and compare the filter characteristic at the normal time with that at the inspection. The threshold for anomaly detection is 0
It may be determined in advance between 1 and 1, or the standard deviation calculated at the time of averaging the filter characteristics under normal conditions may be stored for each frequency component and used as a threshold value reference. Further, the maximum absolute value of the difference between the filter characteristics of the normal signal may be stored for each frequency component and used as a threshold value reference. The threshold is p, which is the threshold reference
(P is a positive real number). Whichever is used, the value is between 0 and 1 and is easy to set.

Next, the operation of abnormality detection will be described with reference to FIG. FIG. 7 is a block diagram showing the principle of abnormality detection according to the present embodiment. 11 is a signal input means, 12 is an adaptive digital filter means, and 13 is an abnormality detection means. The signal input means 11 has a sensor for inputting a one-dimensional time series signal, and is an A / D for converting an analog signal into a digital signal.
The conversion unit includes an anti-aliasing filter for satisfying the sampling theorem when performing A / D conversion. Also,
The sensor that inputs a one-dimensional signal may be any sensor that inputs a one-dimensional time-series signal, such as a microphone or an acceleration sensor. The adaptive digital filter means 12 calculates the filter coefficient W _k from the signal x _k input by the signal input means 11 based on the above equations 4 to 9. In the next abnormality detecting means 13 detects an abnormality of the signal by comparing the filter coefficient W _k obtained.

FIG. 8 is a block diagram showing an example of the abnormality detecting means according to the present embodiment. 21 is a filter characteristic calculating means, 22 is a filter characteristic storing means, 23 is a filter characteristic abnormality detecting means, and 24 is a switch a. , 25 is a switch b, and 26 is a switch c. Next, the operation of the abnormality detecting means 13 will be described. In FIG. 8, switch a24 and switch b2
5. By the ON / OFF of the switch c26, the flow of saving the normal filter characteristics as a database,
Since the flow at the time of inspection for detecting a signal abnormality is different, both will be described. When saving the normal filter characteristics, the switch a24 in FIG.
25 is turned on and switch c26 is turned off. FIG. 9 shows the flow of normal processing in this configuration. A signal is input to the signal input means 11, and the adaptive digital filter means 12 performs a predetermined time adaptive digital filter process on the input signal to obtain a filter coefficient. Next, the filter characteristic calculation means 21 calculates the spectral density distribution of the filter coefficients. Predetermined n for calculation of spectral density distribution (n is an integer of 1 or more)
The above process is repeated until it is obtained. The average value is calculated from the spectral density distribution for n times. Filter characteristic storage means
At 22, the average value is stored as a normal filter characteristic, and a predetermined value is stored as a reference of a threshold value for inspection.

At the time of checking for abnormality, the switch a24 in FIG. 8 is turned on, the switch b25 is turned off, and the switch c26 is turned on.
Let N. Fig. 1 shows the flow of processing at the time of abnormality inspection with this configuration.
It shows in 0. A signal is input to the signal input means 11, and the adaptive digital filter means 12 performs adaptive digital filter processing on the input signal for the same time as in the normal time to obtain a filter coefficient. After this, the filter characteristic calculation means 21
In, the spectral density distribution of the filter coefficient is calculated to obtain the filter characteristic. Next, the filter characteristic abnormality detecting means 23 calculates the absolute value of the difference between the calculated filter characteristic and the normal filter characteristic stored in the filter characteristic storing means 22. Next, it is checked whether or not the absolute value of the difference exceeds the threshold value by using the threshold value stored in the filter characteristic storage means 22 multiplied by p as a threshold value.
p is determined in advance. At this time, if the threshold value is exceeded, it is determined to be abnormal, and if the threshold value is not exceeded, it is determined to be normal.

Although the threshold value reference is stored in the filter characteristic storage means 22 as a predetermined value in FIG. 9, either the standard deviation or the maximum absolute value of the difference is used. Alternatively, two or more may be selected and the abnormality determination may be performed simultaneously. For example, as shown in FIG. 11, the filter characteristic calculation means 21 calculates the average value from the spectral density distribution for n times and also calculates the standard deviation. The filter characteristic storage means 22 stores the average value as a filter characteristic in a normal state, and also stores a standard deviation as a reference of a threshold value during inspection. In this case, the standard deviation stored in the filter characteristic storage means 22 is used as a threshold value to determine whether the filter characteristic abnormality detection means 23 is normal or abnormal. Further, as shown in FIG. 12, the filter characteristic calculation means 21 calculates an average value from the spectral density distributions of n times, and also calculates the maximum absolute value of the difference of each frequency of the n spectral density distributions. The filter characteristic storage means 22 stores the average value as a filter characteristic in a normal state, and also stores the maximum absolute value of the difference for each frequency as a reference of a threshold value for inspection. In this case, the filter characteristic abnormality detecting unit 23 determines whether the filter characteristic is normal or abnormal by using the maximum absolute value of the difference stored in the filter characteristic storing unit 22 as a threshold value.

By utilizing the fact that the absolute value of the difference between the filter characteristics of the filter in the normal state and the filter characteristic in the abnormal state generated by the adaptive digital filter is large, the threshold value for determining the normal state or the abnormal state is set. Easy to set,
Further, it is possible to detect an abnormality even when the abnormality signal is small. The comparison of the filter characteristics for the normal signal and the abnormal signal is performed in the form of spectral density, but the filter coefficient value _Wk may be directly compared.

Embodiment 2. FIG. 13 is a block diagram showing details of the filter characteristic calculation means 21 according to the present embodiment. In the present embodiment, it is checked whether or not the normal signal data for obtaining the filter characteristic at the normal time is appropriate. It is equipped with a means to confirm. Reference numeral 71 is a filter characteristic calculator for calculating filter characteristics, and 111 is a normal time filter characteristic determiner. Figure 1 shows the flow of normal processing in this configuration.
4 shows.

The normal operation will be described below with reference to FIG. Input a signal in the signal input means 11,
In the adaptive digital filter means 12, the input signal is subjected to a predetermined time adaptive digital filter processing to obtain a filter coefficient. The filter characteristic calculator 71 obtains the spectral density distribution of the filter coefficient. The above process is repeated until the spectral density distribution is obtained n times which is determined in advance. After obtaining n times, the average value and standard deviation are obtained from the spectral density distribution of n times. Next, the normal-time filter characteristic determiner 111 checks whether the standard deviation exceeds a predetermined threshold value. If it exceeds, it is considered unsuitable and the above process is repeated. If it does not exceed, the obtained average value is stored in the filter characteristic storage means 22 as being appropriate as the filter characteristic. If necessary, the standard deviation is stored as a threshold standard. The operation at the time of detecting an abnormality is the same as that of the first embodiment, and the standard deviation stored may be used as a reference of the threshold value in the determination of normality or abnormality, or a predetermined value may be used. .

FIG. 15 is a flow chart showing the processing when the threshold value object for judging whether the filter characteristic is appropriate or not is different from the standard deviation shown in FIG. The filter characteristic calculator 71 calculates the spectral density distribution of the filter coefficients obtained by the adaptive digital filter means 12. The spectral density distribution is repeated for a predetermined number of times n, and the average value and the maximum absolute value of the difference between the frequency components of the spectral density distribution for the number n are obtained from the spectral density distribution for the n times. Next, the filter characteristic determiner during normal operation
At 111, it is checked whether the absolute value of the difference between the frequencies exceeds a predetermined threshold value. If it exceeds, it is considered unsuitable and the above process is repeated. If it does not exceed, the obtained average value is stored in the filter characteristic storage means 22 as being appropriate as the filter characteristic. If necessary, the maximum absolute value of the difference is stored as a threshold value reference. The operation at the time of detecting an abnormality is the same as that of the first embodiment, and the maximum absolute value of the stored difference may be used as the threshold value criterion in determining whether the abnormality is normal or a predetermined value. May be used. Any method may be used as the method for determining the filter characteristics in the normal state shown in FIGS. 14 and 15, or two methods may be used at the same time.

In this way, the filter characteristic calculating means 21 checks whether or not it is appropriate as a filter characteristic in a normal state, and adopts the filter characteristic when it is appropriate, thereby improving the reliability as a filter characteristic in a normal state. Is improved.

Embodiment 3. FIG. 16 is a block diagram showing details of the filter characteristic storage means 22 according to the present embodiment. In the present embodiment, a plurality of normal filter characteristics are stored in the filter characteristic storage means 22 for inspection. It is provided with a means for selecting a filter characteristic according to the situation of time. Reference numeral 72 is a filter characteristic memory for storing the characteristics of the filter, 121 is a status memory, and 122 is a filter characteristic selector.

The normal filter characteristic obtained by the filter characteristic calculating means 21 is stored in the filter characteristic storage 72. When the normal filter characteristics are saved, the status in which the normal filter characteristics are obtained is saved in the status storage unit 121 at the same time. The situation is, for example, measurement position, temperature, humidity,
There are various situations such as the season, time, plant operating status, and the age of equipment. The normal filter characteristics to be saved are prepared for various situations.

Next, at the time of abnormality inspection, the current situation is obtained. Then, by the filter characteristic selector 122, the situation memory 121
The situation stored in is compared with the current situation, and the filter characteristic calculated in the same situation as the current situation is selected. That is, the optimum normal filter characteristic is selected and sent to the filter characteristic abnormality detecting means 23.

In this way, a plurality of normal filter characteristics are stored in the filter characteristic storage means 22 according to the situation, and the normal filter characteristic corresponding to the situation at the time of abnormality detection is selected to obtain the optimum filter characteristic. Will be used, and the anomaly detection performance will be improved.

Fourth Embodiment FIG. 17 is a block diagram showing details of the filter characteristic abnormality detecting means 23 according to the present embodiment. In the present embodiment, a means for checking and confirming whether or not the filter characteristic at the time of abnormality detection is appropriate is provided. I have it. Reference numeral 81 is a filter characteristic abnormality detector for calculating abnormality of the filter characteristic, and 131 is an abnormality detection determinator. FIG. 18 shows the flow of processing when abnormality is detected in this configuration.

In the process shown in FIG. 18, the abnormality detection determinator 13
In 1), the abnormality detection is finally confirmed based on multiple consecutive abnormality detections. The details of this processing will be described.
An abnormality determination is made in the filter characteristic abnormality detector 81,
The determination content is sent to the abnormality detection determining device 131. In the abnormality detection determiner 131, the total number of determinations made by inspection and the number of determinations made as abnormal by inspection are accumulated. Then, it is checked whether or not the accumulated number of determinations has reached m, and if the number of determinations is less than m, the process is returned to the signal input means 11. When the number of determinations reaches m, it is checked whether or not the number of next abnormal determinations exceeds k. If the number of times of determination as abnormality does not exceed k times within the predetermined determination number m, it is determined as normal, and if it exceeds, determination of abnormality is confirmed.

FIG. 19 shows another abnormality determination process of the abnormality detection / determination unit 131. This is a flow chart of processing for determining abnormality detection when the number of consecutive abnormality determinations exceeds a predetermined h times. An abnormality determination is made in the filter characteristic abnormality detector 81, and the determination content is the abnormality detection determinator 13
Sent to 1. The abnormality detection determiner 131 checks whether the judgment is abnormal. If there is no abnormality, the number of continuous abnormality determinations is reset to 0. In the case of abnormality determination, the number of continuous abnormality determinations is accumulated. After that, it is checked whether the number of continuous abnormality determinations exceeds a predetermined h times. If not exceeded, the process is returned to the signal input means 11, and if exceeded, the abnormality detection is confirmed. Normal while abnormal detection is not confirmed.

As described above, the filter characteristic abnormality detecting means 23 finally determines the abnormality detection based on the plurality of times of abnormality detection, thereby reducing false alarms and highly reliable abnormality detection. The abnormality detection method is shown in FIG. 18 and FIG.
Any of 9 methods may be used, or two may be used simultaneously.

Embodiment 5. FIG. 20 is a block diagram showing an example of the abnormality detecting means according to the present embodiment. 31 is a filter coefficient storage means, 32 is a filter cut means, 33 is a filter cut component storage means, and 34 is a filter cut component abnormality detection means. , 35 is a switch d, 36 is a switch e, and 37 is a switch f.

Next, the procedure for storing the normal filter coefficient will be described. The switch d35 shown in FIG. 20 is turned off, the switch e36 is turned off, and the switch f37 is turned on. Signal input means to obtain the normal filter coefficient
The signal input from 11 is applied to the adaptive digital filter means 12
To obtain the filter coefficient. The obtained normal filter coefficient is stored in the filter coefficient storage means 31.

Next, the processing at the time of abnormality inspection will be described.
Switch d35 shown in FIG. 20 is turned on and switch e36 is turned on.
N, switch f37 is turned off. FIG. 21 is a flow chart showing the flow of processing at the time of abnormality detection. From the signal input from the signal input means 11, the normal time filter coefficient stored in the filter coefficient storage means 31 is used by the filter cut means 32 to remove the normal time component. A signal obtained by removing the normal component from the input signal is used as a filter cut component.

The filter cutting means 32 will be described below. FIG. 22 is a diagram for explaining the filter cutting means 32, and 181 is a digital filter processing section. In the digital filter processing section 181, the input signal x _k is convoluted using the filter coefficient in the normal state stored in the filter coefficient storage means 31 to become the output y _k containing only the component in the normal state. On the other hand, the input signal x _k becomes the target signal d _k which is delayed by the delay device 171 by about ½ of the filter length. The target signal d _k and the output y are added by the adder 162.
Signals of which the sign of _k is inverted are added, and the target signal d _k
A filter cut component, which is a signal _obtained by subtracting the output y _k from, is obtained. This filter cut component becomes a signal from which the component in the normal state has been removed.

The filter cut component obtained by the filter cut means 32 as described above is always stored in the filter cut component storage means 33 for a certain past time as shown in FIG. Filter cut component abnormality detection means 34
Then, the abnormality is detected from the filter cut component by two methods. First, the standard deviation of the filter cut components stored for a certain period of time is calculated and compared with a threshold value.
This threshold value may be determined in advance or may be set based on the standard deviation during normal operation. As a result of comparison,
If it exceeds the threshold value, an abnormality is detected, and if it does not exceed the threshold value, it is determined to be normal. The other method is to compare the filter cut component itself with a threshold value, detect an abnormality if it exceeds, and determine normal if it does not exceed. The threshold value in this case may be a predetermined value, or may be set based on the standard deviation during normal operation. If either one of the above two methods detects an abnormality, the abnormality is detected.

As described above, by using the standard deviation of the filter cut components, it is possible to judge the abnormality by the components excluding the normal components. Moreover, since the filter cut component is a real-time signal, it is possible to detect a sudden abnormal signal.

Sixth Embodiment FIG. 23 is a block diagram showing details of the filter coefficient storage means 31 according to the present embodiment. In the present embodiment, a plurality of normal filter characteristics are stored in the filter coefficient storage means 31 for inspection. It is provided with a means for selecting a filter characteristic according to the situation of time. Reference numeral 91 is a filter coefficient storage for storing the filter coefficients, 121 is a status storage, and 141 is a filter coefficient selector.

The normal-time filter coefficient obtained by the adaptive digital filter means 12 is stored in the filter coefficient memory 91. When saving the normal filter coefficient,
The situation in which the filter coefficient at the normal time was obtained is stored in the situation storage unit 121.
Save at the same time. When the normal filter characteristics are saved, the status in which the normal filter characteristics are obtained is saved in the status storage unit 121 at the same time. The status includes various statuses such as measurement position, temperature, humidity, season, time, operating status of equipment to be inspected, and number of years elapsed. A plurality of normal-time filter coefficients to save are prepared for all situations.

Next, at the time of abnormality inspection, the current situation is obtained. Then, by the filter coefficient selector 141, the situation memory 121
The situation stored in is compared with the current situation, and the filter characteristic calculated in the same situation as the current situation is selected. That is, the optimum normal filter characteristic is selected and sent to the filter cutting means 32.

As described above, by storing a plurality of normal-time filter characteristics in the filter coefficient storage means 31 according to the situation and selecting the normal-time filter coefficient optimum for the situation at the time of abnormality detection, the abnormality detection performance is improved. improves.

Seventh Embodiment FIG. 24 is a block diagram showing details of the filter cut abnormality detecting means 34 according to the present embodiment. In the present embodiment, means for checking and confirming whether or not the filter cut component at the time of abnormality detection is appropriate. Is equipped with. Reference numeral 101 is a filter cut component abnormality detector for calculating abnormality of the filter cut component, and 131 is an abnormality detection determinator. The flow of processing at the time of abnormality detection in this configuration is the same as that shown in FIGS. However,
The filter cut component abnormality detector 101 is not applied when a sudden abnormality signal is detected.

That is, the filter cut component abnormality detecting means 34
In FIG. 18 also, as shown in FIG. 18, the abnormality detection is finally determined when the number of times of determination of abnormality out of the total number of inspection times exceeds a predetermined number. Further, as shown in FIG. 19, the abnormality detection is finally determined based on a plurality of consecutive abnormality determinations. In this way, even when detecting anomalies due to filter cut components, by finally determining anomaly detection based on anomaly detection multiple times, false alarms are reduced,
It is possible to perform highly reliable abnormality detection.

Eighth Embodiment FIG. 25 is a block diagram showing an example of the abnormal signal source detection device according to the present embodiment,
Reference numeral 51 is an abnormal signal source detecting means. The signal input means 11 obtains an input signal based on at least two different sensors.

FIG. 26 is a flow chart for explaining the processing of the abnormal signal source detecting means 51 according to this embodiment. When an abnormality is detected by the filter cut component abnormality detection means 34, the abnormal signal source detection means 51 determines the time difference between the input sensors from the filter cut components of the plurality of input sensors stored in the filter cut component storage means 33. To calculate. Then, the position of the abnormal signal source is calculated based on the equations 10 to 19.

At the time of abnormality, the filter cut component contains almost no normal signal component, but contains much abnormal signal component. Therefore, from the filter cut components obtained from two or more input sensors, An abnormal signal source can be detected. In the above description, for example, sound is used as the signal source, but a signal other than sound may be used as long as the transmission speed is constant in the medium that transmits the signal, and the input sensor may be one that can input the input signal.

Ninth Embodiment FIG. 27 is a block diagram showing the configuration of the abnormality detecting device according to the present embodiment. In this embodiment, the abnormality detection based on the filter characteristic and the abnormality detection based on the filter cut component are performed, and when at least one of them is determined to be abnormal, it is determined that the abnormality is detected. That is, the abnormality detecting means 13 is performed by the filter characteristic abnormality detecting means 23 in the first embodiment and the filter cut component abnormality detecting means 34 in the fifth embodiment.
In this way, the abnormality detection performance is improved by simultaneously performing the abnormality detection based on the filter characteristic and the abnormality detection based on the filter cut component.

Embodiment 10. FIG. 28 is a block diagram showing the configuration of the abnormality detection / abnormality signal source detection device according to the present embodiment. This embodiment has both the abnormality detecting device according to the ninth embodiment and the abnormality signal source detecting device according to the eighth embodiment. The abnormality detection is performed by the filter characteristic abnormality detection means 23 and the filter cut component abnormality detection means 34, and when abnormality is detected in at least one of them, the abnormality signal source detection means 51 determines the distance between the plurality of input sensors. An abnormal signal source is detected. By detecting the abnormal signal source even when an abnormality is detected by the filter characteristic in this way, the performance of the abnormality detection and the abnormal signal source detection is improved.

[0100]

The abnormal signal detecting apparatus according to the first aspect of the present invention comprises an adaptive digital filter for simulating the characteristics of a delay device for an input signal which is a one-dimensional time series signal,
An abnormal signal is detected by comparing the filter characteristic in the filter characteristic forming process of the adaptive digital filter with respect to the normal input signal and the filter characteristic in the filter characteristic forming process of the adaptive digital filter with respect to the abnormal input signal. Since the abnormality detection means is provided, there is an effect that an abnormality signal detection device capable of detecting an abnormality can be obtained even when the component of the abnormality signal is small.

According to the second structure of the present invention, the first
In the above configuration, a normal time filter characteristic judging means for judging whether or not the filter characteristic in the filter characteristic forming process of the adaptive digital filter with respect to a normal time input signal is proper, and the abnormality detecting means compares As the filter characteristic, the filter characteristic determined by the normal-time filter characteristic determination unit is determined to be used, thereby improving the reliability of the normal-time filter characteristic to be compared and improving the abnormality detection performance. There is an effect that an abnormal signal detection device that can be improved can be obtained.

According to the third structure of the present invention, the first
In the second configuration, a filter characteristic storing means for storing the filter characteristic in the filter characteristic forming process of the adaptive digital filter with respect to the input signal in the normal state is provided, and the normal-time filter characteristic to be compared by the abnormality detecting means is the above-mentioned. Since the filter characteristic stored in the filter characteristic storage means is used, there is an effect that an abnormal signal detecting device capable of quickly detecting an abnormality can be obtained.

According to the fourth aspect of the present invention, the first
Further, in the second configuration, a filter characteristic storage means for storing a plurality of filter characteristics for normal input signals according to the situation is provided, and the filter characteristic storage means is used as a normal filter characteristic to be compared by the abnormality detection means. The abnormal signal detection device that can detect abnormalities by following changes in the situation by improving the abnormal performance by configuring the filter characteristics that are close to the situation at the time of abnormality detection among the multiple filter characteristics saved in There is an effect to be obtained.

According to the fifth aspect of the present invention, the adaptive digital filter for simulating the characteristics of the delay device with respect to the input signal which is a one-dimensional time-series signal, and the adaptive digital filter with respect to the normal input signal are provided. Filter coefficient storage means for storing a filter coefficient in the filter characteristic forming process of the filter, and filter cut means for obtaining a filter cut component obtained by cutting the normal component from the input signal at the time of abnormality detection by using the normal filter coefficient Since the abnormal signal detecting means for detecting the abnormal signal based on the filter cut component which is the output signal of the filter cutting means is provided, there is an effect that an abnormal signal detecting device capable of detecting a sudden abnormal signal can be obtained.

According to the sixth aspect of the present invention, the fifth
In the above configuration, a filter coefficient storage means for storing a plurality of filter coefficients for normal input signals according to the situation is provided, and the filter characteristic storage means is used as a normal-time filter coefficient to be used for cutting by the filter cutting means. By configuring to use the filter coefficient of the situation close to the situation at the time of abnormality detection among the plurality of filter coefficients stored in, it is possible to detect a sudden abnormal signal,
There is an effect that an anomaly signal detecting device can be obtained which can detect anomalies following changes in circumstances and improve anomaly performance.

According to the seventh aspect of the present invention, the first
In any one of the sixth to sixth configurations, since the abnormality detecting means outputs the abnormality detection signal when the abnormality is detected a plurality of times consecutively, false alarms can be reduced and reliability is high. There is an effect that the abnormal signal detection device can be obtained.

According to the eighth aspect of the present invention, the fifth
In the configuration of, further comprising a plurality of sensors, and means for storing a plurality of filter cut components of the same time zone based on the plurality of sensors, when an abnormal signal is detected,
The position of the abnormal signal source is calculated by calculating the time difference between the filter cut components based on the plurality of sensors, so that the abnormal signal source can be detected using the filter cut component and the position of the abnormal signal source can be detected. There is an effect that the detection device can be obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an adaptive digital filter according to a first embodiment of the present invention.

FIG. 2 is a graph relating to the first embodiment and showing a filter characteristic by a spectral density with respect to a frequency in a process of generating a filter that passes a normal signal by adaptive digital filter processing.

FIG. 3 is a graph relating to the first embodiment and superposing three filter characteristics after performing adaptive digital filter processing for 2 seconds in different time zones of a normal signal.

FIG. 4 is a graph relating to the first embodiment, in which the filter characteristics of the filters obtained by performing the adaptive digital filter processing on the normal signal and the abnormal signal at the same time are overlapped.

FIG. 5 is a graph relating to the first embodiment and showing an absolute value of a difference in filter characteristics after performing adaptive digital filter processing for 2 seconds on the normal signal and the abnormal signal of FIG. 4;

FIG. 6 is a graph relating to the first embodiment and superimposing three absolute values of differences in filter characteristics after performing adaptive digital filtering for 2 seconds on the three normal signals shown in FIG.

FIG. 7 is a block diagram showing the principle of detecting an abnormal signal according to the first embodiment.

FIG. 8 is a block diagram showing an example of abnormality detecting means according to the first embodiment.

FIG. 9 is a flowchart illustrating processing of a normal signal according to the first embodiment.

FIG. 10 is a flowchart illustrating processing when abnormality is detected according to the first embodiment.

FIG. 11 is a flowchart illustrating another process of the normal signal according to the first embodiment.

FIG. 12 is a flowchart illustrating another processing of the normal signal according to the first embodiment.

FIG. 13 is a block diagram showing details of a filter characteristic calculation means 21 according to the second embodiment of the present invention.

FIG. 14 is a flowchart illustrating processing of a normal signal according to the second embodiment.

FIG. 15 is a flowchart illustrating another process of the normal signal according to the second embodiment.

FIG. 16 is a block diagram showing details of a filter characteristic storage means 22 according to a third embodiment of the present invention.

FIG. 17 is a block diagram showing details of a filter characteristic abnormality detecting means 23 according to a fourth embodiment of the present invention.

FIG. 18 is a flowchart illustrating processing of the abnormality detection determinator 131 according to the fourth and seventh embodiments.

FIG. 19 is a flowchart illustrating another process of the abnormality detection determinator 131 according to the fourth and seventh embodiments.

FIG. 20 is a block diagram showing an example of abnormality detecting means according to a fifth embodiment of the present invention.

FIG. 21 is a flowchart showing the flow of processing at the time of abnormality detection according to the fifth embodiment.

FIG. 22 is a filter cutting means according to the fifth embodiment.
FIG. 33 is a block diagram illustrating 32.

FIG. 23 is a block diagram showing details of a filter coefficient storage means 31 according to a sixth embodiment of the present invention.

FIG. 24 is a block diagram showing details of filter cut abnormality detection means 34 according to a seventh embodiment of the present invention.

FIG. 25 is a block diagram showing an example of an abnormal signal source detection device according to an eighth embodiment of the present invention.

FIG. 26 is an abnormal signal source detection means according to the eighth embodiment.
It is a flow chart explaining processing of 51.

FIG. 27 is a block diagram showing a configuration of an abnormality detection device according to a ninth embodiment of the present invention.

FIG. 28 is a block diagram showing a configuration of an abnormality detection / abnormal signal source detection device according to a tenth embodiment of the present invention.

FIG. 29 is a waveform showing a normal signal of sound at a normal time in the plant.

FIG. 30 shows the normal signal of FIG. 29 for every 256 points.
It is a graph which shows the spectrum density distribution averaged after performing four times Fourier transform.

FIG. 31 is a waveform showing an abnormal signal in which an abnormal sound is added to a normal sound in a plant.

FIG. 32 shows the abnormal signal of FIG. 31 at 100 points every 256 points.
It is a graph which shows the spectrum density distribution averaged after performing four times Fourier transform.

33 is a graph showing a spectral density distribution obtained by Fourier-transforming the abnormal sound added in FIG. 31 100 times for every 256 points and averaging it.

34 is a graph showing the absolute value of the difference between the spectral density distributions of FIGS. 30 and 32 and twice the standard deviation of the normal signal of FIG. 30. FIG.

FIG. 35 is a block diagram showing a basic configuration of a conventional adaptive digital filter.

FIG. 36 is an explanatory diagram illustrating a conventional method for detecting an abnormal signal source from two filter cut components.

FIG. 37 is an explanatory diagram illustrating a conventional method for detecting an abnormal signal source from three filter cut components.

FIG. 38 is an explanatory diagram illustrating a conventional method for detecting an abnormal signal source from four filter cut components.

[Explanation of symbols]

11 signal input means, 12 adaptive digital filter means,
13 abnormality detecting means, 21 filter characteristic calculating means, 22 filter characteristic storing means, 23 filter characteristic abnormality detecting means, 31
Filter coefficient storage means, 32 Filter cut means, 33
Filter cut component storage means, 34 Filter cut component abnormality detection means, 51 Abnormal signal source detection means, 71 Filter characteristic calculator, 72 Filter characteristic storage, 81 Filter characteristic abnormality detector, 91 Filter coefficient storage, 101 Filter cut component Abnormality detector, 111 Normal filter characteristic determiner, 121 Situation memory, 122 Filter characteristic selector, 131 Abnormality detection determiner, 141 Filter coefficient selector, 161 Adaptive digital filter processing unit, 162 Adder, 171 Delay device, 181 Digital filter processing unit,
291a, 291b, 291c, 291d microphones.

Continuation of the front page (56) References JP-A-5-27782 (JP, A) JP-A-6-3215 (JP, A) JP-A-6-1995094 (JP, A) Actual Kaihei 4-75498 (JP , U) Katsutomo Hanakuma, et al., Application of Complex Process Abnormal Signal Detection System in Petrochemical Plant, Instrumentation, Japan, Kogyo Gijutsu, September 1, 1995, Volume 38, No. 9, 36 −40 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G05B 23/02 302 G01H 17/00 H03H 17/02 H03H 21/00

Claims (8)

(57) [Claims] 1. An adaptive digital filter for simulating the characteristics of a delay device with respect to an input signal which is a one-dimensional time-series signal, and a filter characteristic in a filter characteristic forming process of the adaptive digital filter with respect to a normal input signal, An abnormal signal detecting device comprising an abnormality detecting means for detecting an abnormal signal by comparing a filter characteristic in the process of forming the filter characteristic of the adaptive digital filter with respect to an input signal at the time of abnormality. 2. A normal time filter characteristic judging means for judging whether or not a filter characteristic in the filter characteristic forming process of the adaptive digital filter with respect to a normal time input signal is proper, and a normal time for comparing with the abnormality detecting means. 2. The abnormal signal detecting apparatus according to claim 1, wherein the filter characteristic determined by the normal-time filter characteristic determining means is used as the filter characteristic of 1. 3. A filter characteristic storing means for storing a filter characteristic in a process of forming a filter characteristic of the adaptive digital filter with respect to an input signal in a normal state, wherein the filter characteristic is a normal time filter characteristic to be compared by the abnormality detecting means. The abnormal signal detection device according to claim 1 or 2, wherein the filter characteristic stored in the storage means is used. 4. A filter characteristic storing means for storing a plurality of filter characteristics for a normal input signal according to the situation, and storing in the filter characteristic storing means as a normal filter characteristic to be compared by the abnormality detecting means. The abnormal signal detecting device according to claim 1 or 2, wherein a filter characteristic of a situation close to a situation at the time of abnormality detection is used among the plurality of filter characteristics. 5. An adaptive digital filter for simulating characteristics of a delay device for an input signal which is a one-dimensional time-series signal, and a filter coefficient in a process of forming a filter characteristic of the adaptive digital filter for a normal input signal is stored. Filter coefficient storage means, filter cut means for obtaining a filter cut component obtained by cutting a normal component from an input signal at the time of abnormality detection by using the normal filter coefficient, and a filter which is an output signal of the filter cut means. An abnormality signal detecting device, comprising: an abnormality detecting unit that detects an abnormal signal based on a cut component. 6. A filter coefficient storage means for storing a plurality of filter coefficients for a plurality of normal-time input signals according to the situation, wherein the filter characteristic storage is provided as a normal-time filter coefficient used when the filter cutting means cuts. The abnormal signal detecting apparatus according to claim 5, wherein the filter coefficient stored in the means is the filter coefficient in a situation close to the situation at the time of abnormality detection. 7. The abnormality detecting means is configured to output an abnormality detection signal when the abnormality is detected a plurality of times in succession. Abnormal signal detection device. 8. The apparatus according to claim 5, further comprising a plurality of sensors, and means for storing a plurality of filter cut components in the same time zone based on the plurality of sensors, the plurality of sensors being provided when an abnormal signal is detected. An abnormal signal source detection device, wherein the position of the abnormal signal source is calculated by calculating the time difference between the filter cut components based on the sensor.