JPH027409B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a device that records the state of a plant and, if necessary, determines the cause of an abnormality based on the recorded data, and compresses and records plant data information. The present invention also relates to a diagnostic device that detects the occurrence of an abnormality and determines the cause of the abnormality by referring to recorded characteristics.

Conventionally, many methods have been used to diagnose abnormalities and failures in large-scale plants. Based on the plant signal actually measured in this one,
Prepare information in advance to deal with abnormalities and breakdowns,
There is a method of diagnosing abnormalities and failures by comparing plant signals with this characteristic information. For this purpose, it is of course necessary to record operating data before and after the occurrence of an abnormality, but also to record operating data during normal times in order to distinguish between an abnormality in which the operating state changes gradually and changes over time in the plant. It is necessary to record at appropriate time intervals. However, in large-scale plants, the number of signal points required to record can reach several hundred, and data from such a large number of points cannot be recorded.
Continuous recording requires a huge amount of storage capacity.
This problem arises. In order to solve this problem, when the increase or decrease in operational data during measurement exceeds a certain rate, the data in a certain section immediately before that is recorded at a short sampling period, and the data even earlier is recorded. A plant data recording method has been proposed that prevents an explosive increase in recorded data by lengthening the sampling period for recording as time passes and discarding data between samples ( Nagaoka et al., development of abnormal plant data record retrieval equipment, thermal and nuclear power generation,
vol.32, No.11, P1169).

However, in the above recording method, the sampling period is not constant, making statistical analysis of the measured data difficult, and spectral components cannot be used as characteristic information when diagnosing abnormalities or failures. That is, although this spectrum is very effective as information expressing the vibrational properties due to abnormalities occurring in the plant, the above method has the disadvantage that it is not possible to adopt a diagnostic method based on this information. Furthermore, in the above method, abnormality detection is determined based on whether the measured signal from the plant exceeds a certain value.
In this method, when setting a certain value, it is not clear how to set it to detect an abnormality early and reliably. In this way, the above method has the drawback that by changing the sampling period, the spectral characteristics of the data are lost, which limits the diagnostic method for diagnosing abnormalities and accidents, and also makes it difficult to detect abnormalities. Even in law, ambiguity remains.

The purpose of the present invention is to use measurement signals from a plant to diagnose abnormalities and accidents in large-scale plants.
To provide a device that compresses and records information so that the spectral structure can be maintained, constantly verifies the occurrence of an abnormality, and when an abnormality occurs, identifies the cause based on the recorded information. be.

In order to achieve the above object, the present invention includes a digital filter device that stores various digital filter characteristics and converts digitally converted plant data; and a digital filter device that stores various digital filter characteristics and converts digitally converted plant data; a signal device that outputs a signal to the digital filter device to determine a filter error based on the retrieved digital filter characteristics; a verification device that verifies the steady state and unsteady state of the plant based on the error; A recording device for recording information including digital filter characteristics corresponding to a state and an unsteady state, and a recording device for verifying whether the unsteady state is due to artificial plant operation and control when the unsteady state of the plant is verified. an abnormality testing device, and a spectrum calculation device that calculates spectra corresponding to the steady state and unsteady state of the plant based on the digital filter characteristics recorded in the recording device, using the previously recorded spectrum and the spectrum calculation device. The system is equipped with an abnormality determination device that compares the obtained spectrum with the obtained spectrum and determines whether there is an abnormality in the plant.

In the present invention, first, real-time verification of stationarity is performed using a digital filter. Generally, in a digital filter, if a certain measurement signal from the plant at time i is expressed as xi, then the filter output yi is yi= _p 〓 ^k=0 b _k x _ik + _d 〓 ^k=1 a _k y _ik ……(1) It has a structure that can be expressed as Filter parameters b _k , (k=0, 1, ..., p), a _k , (k
=1, 2, . . . , q) is a parameter that can maintain the dynamic structure of the data and therefore also the spectral structure, and this value can be obtained offline or online as described later. Further, the above-mentioned integers p and q each have a value of about 10 at most, and these can also be determined in advance. For example, in the above digital filter, if the filter error ei is ei=xi−{ _p 〓 ^k=0 b _k x _ik + _d 〓 ^k=1 a _k y _ik } …(2), then this variance σ ² _e = 1/N _N 〓 ¹ e ² i ...(3) is calculated, and the information criterion AIC as the evaluation quantity of model validity AIC=-2logσ ² _e +2 (p+q) ...(4) is minimized. It is known that the integers p and q can be determined once in advance (Soeda et al., Basics and Applications of Signal Processing, Nissin Publishing).

When using such a digital filter, for example, the number of samples is about 1000 to 5000, that is, if samples are sampled at 1 second intervals, data from 1000 to 5000 seconds will be represented by about p + q (~20) numbers. In the above example, the information compression effect is approximately 2
It can be seen that it reaches 0/1000 to 20/5000.

By the way, the filter error ei expressed by equation (2) is an amount obtained from moment to moment. If the plant, which is currently in a steady state, changes due to some abnormality, the measurement signal xi will change, but the filter output will not be able to instantaneously respond to the change, so the error ei will increase. Therefore, by constantly monitoring the error ei, the occurrence of an abnormality can be detected at an early stage.

Furthermore, the filter characteristics expressed by parameters {a _k } and {b _k } are converted into the spectral domain as follows: This shows the spectral structure of the measurement signal xi from the plant. In this way, it is shown that by applying a digital filter, information compression of plant data can be effectively performed, and it is also highly effective in detecting abnormalities.

Hereinafter, the present invention will be explained with reference to Examples. 1st
The figure shows an example of a plant status recording/monitoring device according to the present invention. The signal 16 measured in the plant 1 is processed by the input device 2 by removing the bias included in the signal 16, sampling at predetermined time intervals, analog-to-digital conversion, and combining closely related plant signals. Signals are divided into groups, etc. The grouped digital signals are sent to the signal device 3. The signal device 3 has a protocol for synchronizing the grouped digital signals with the digital filter characteristics corresponding to the signals, and sequentially synchronizes the corresponding digital filter characteristics stored in the digital filter device 4. The characteristics are called and the error signal 18 is determined. At this time, the digital filter device 4 uses the formula
It is also possible to obtain a filter having a structure like (1) in real time, and FIG. 2, which will be described later, explains the structure of this digital filter device.

Next, the test device 5 performs a test of stationarity using the error signal 18. To test this, for example, calculate α=√ ² from the variance σ ² of the time series regarding the error ei in a steady state, and consider the point at which |ei|≧3α ...(6) as the point at which it ceases to be stationary. be able to.

When stationarity is established, the filter characteristics, that is, the compressed plant information, the error value ei at that time, and the average value i of the plant signal are sent from the filter device 4 to the recording device 6 in correspondence with each plant signal. It is captured. Signal 19 is recording device 6
This shows the control signal for importing into the . When stationarity is established, actual recording may be performed at predetermined intervals (1000 to 5000 seconds in the example already described).

When the stationarity deviates, the plant signal xi itself is also immediately taken into the recording device 6 in addition to the above-mentioned filter characteristics. When the operating state of a plant suddenly loses its stability, it can be divided into cases where an abnormality occurs in the plant, and cases where the plant state becomes transient due to human operation or control. In the latter case, there are no problems with plant operation and therefore it is not necessary to activate the diagnostic function. Therefore, in the abnormality detection device 8,
When non-stationarity is detected, it is verified whether it is due to artificial plant operation or control. For this purpose, plant operation signals and control signals are input into the input device 7, and the signals are similarly preprocessed and sampled.
In particular, an averaging operation is performed, and the variance value σ ² _d of each signal is determined. Here, the operation signal or control signal corresponding to the pre-corresponding plant signal similarly becomes relatively large compared to the variance value σ ² _d at the time when non-stationarity is detected, and the average value becomes If there is a certain trend,
It is determined that the verified non-stationarity is due to artificial operation and control of the plant. In other cases, the unsteadiness can be considered to be due to some plant abnormality, and the abnormality is immediately diagnosed.

The signal 20 is a control signal for diagnosis, and the filter characteristics recorded corresponding to the related signal groups are sequentially called from the recording device 6, and the formula (5)
A spectrum calculation device 9 is prepared which calculates a spectrum using a spectrum calculation formula according to the following. Of course, at this time, spectrum calculation is also performed based on the filter characteristics at the time when it is determined to be abnormal. Each spectral density is input to the abnormality determination device 10. Here, a pattern recognition method is used to identify and judge the abnormality. For this reason, when recording on a recording device, the correspondence between fixed abnormalities and recorded data is always made.

For example, the abnormality determination device 10 performs the following calculation. Assume that a spectrum is being determined corresponding to a certain abnormality δ. At this time, the spectrum is sampled n times for each frequency, and a pattern vector X is created whose elements are the sizes of the spectra.
If the current pattern vector in which an abnormality is detected is X, then for example, d ₁ = 1/2 _N 〓 ¹ (X-X〓) ^l _i (X + X〓) ^m _i ...(7) d ₂ = 1/2 _o 〓 ¹ (X−X〓) ^l _i (X・X〓) ^m _i ……(8) d ₃ =1/2 _o 〓 ¹ (X/X〓) ^l _i (X+X〓) ^m _i ……(9 ) is defined, and records corresponding to the abnormality δ are sequentially called from the recording device 6 so as to minimize this distance. In the above formulas (7) to (9), l and m are integers, and ( ) i represents a vector element. The anomaly that minimizes the distance calculated in this way is considered to be the currently occurring anomaly. The abnormality determined in this way is recorded again together with the corresponding filter characteristics.

Now, the abnormality identified in this way is displayed on display device 1.
1, and instructions for abnormality recovery are given to the plant via the man-machine device 12.

FIG. 2 shows an example of the configuration of a digital filter device. In this case, measurement signal 2
1 and a corresponding control signal 22 are given. These signals are part of the signal 17 mentioned above. Now, in response to a certain signal, storage device 2
The digital filter characteristics such as {a _k }, {b _k }, p, q, etc. stored in 3 are recalled. measurement signal di
is input to the filter via the m-stage delay circuit 25. The characteristic of digital filter is filter error ei
can be sought to minimize. The determined filter characteristics are stored again in the storage device 23, and the next measurement signal is input to the digital filter.

Now, as a method for determining the digital filter characteristics of the filter structure given by equation (1) as described above, there is the following method (Soeda et al., Fundamentals and Applications of Signal Processing, Nissin Publication).

a _k,i+1 =a _k,i +ueiα _k,i ...(10) b _k,i+1 =b _k,i +veiβ _k,i ...(11) However, u>0, v>0 α _k,i = ∂yi/∂a _k,i β _k,i = ∂yi/∂b _k,i At the stage when k,i converges, the following formula holds a _k = a _k,i+1 ……(12) b _k =b _k,i+1 ...(13) From the above, the characteristics of the digital filter are
In other words, the dynamic characteristics of the data are determined.

As described above, the present invention enables compressed recording of data information while preserving the spectral structure of plant data, which was difficult with conventional data recording/monitoring devices. This filter characteristic makes it possible to detect an abnormality early and to identify and diagnose the abnormality using a spectrum.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing a plant condition recording and monitoring device of the present invention, and FIG. 2 is a configuration diagram of a digital filter device constituting the device. 3...Signal device, 4...Digital filter device,
5... Verification device, 6... Recording device, 8... Abnormality testing device, 9... Spectrum calculation device, 10... Abnormality determination device.

Claims (1)

[Claims] 1. A digital filter device that stores various digital filter characteristics and converts digitally converted plant data, calls out the digital filter characteristics corresponding to the plant data, and calculates a filter error based on the retrieved digital filter characteristics. A signal device that outputs a desired signal to the digital filter device, a verification device that verifies the steady state and unsteady state of the plant based on the error, and a digital filter characteristic that corresponds to the steady state and unsteady state of the plant. a recording device for recording information contained in the plant; an abnormality verification device for verifying whether or not the unsteady state is due to artificial plant operation and control when the unsteady state of the plant is verified; a spectrum calculation device that calculates spectra corresponding to the steady state and unsteady state of the plant based on digital filter characteristics; A plant condition recording monitoring device including an abnormality determination device for determining.