KR20230102748A - Apparatus for scrubber fan monitoring using adaptive kalman filter and method thereof - Google Patents

Abstract

The present invention relates to a scrubber fan monitoring apparatus using an adaptive Kalman filter and a method thereof, which can improve the accuracy of fault detection. According to the present invention, the scrubber fan monitoring apparatus using an adaptive Kalman filter comprises: a storage part storing raw data collected from a sensor; an observation data generation part using the stored raw data to generate observation data; a Kalman filtering part inputting the observation data into an adaptive Kalman filter to output estimation data estimating observation data with high-frequency noise removed therefrom; a feature value extraction part using the estimation data estimated by the adaptive Kalman filter to extract normalized statistical data as feature values; and a fault determination part counting the number of times the extracted feature values exceed a preset threshold value during a set time to determine the presence of a fault of the scrubber fan.

Description

Scrubber fan monitoring device and method using an adaptive Kalman filter

The present invention relates to an apparatus and method for monitoring a scrubber fan using an adaptive Kalman filter, and more particularly, to an adaptive Kalman filter for extracting feature values in a time domain and diagnosing whether or not a scrubber fan has failed using the adaptive Kalman filter. It relates to a scrubber fan monitoring device using a filter and a method therefor.

Since the environment of the ozone layer is protected by the Clean Air Conservation Act in Korea, failure to comply with these environmental certifications may result in legal and economic disadvantages. Therefore, it is essential that all industrial sites have air purifiers to purify harmful gases. It is becoming.

Therefore, in all industrial sites, dust collectors, adsorption towers, scrubbers, etc. are installed as air dust prevention facilities, and each of these is installed in consideration of efficiency according to the type of pollutant.

Among them, the scrubber has been developed in various forms as a device for deodorizing and purifying contaminated air that generates a high concentration of odor or contains harmful gases. These scrubbers mainly use the principle of deodorizing contaminated air by contacting dispersed liquid particles or deodorizing by using an adsorbent.

As such, it is important to monitor the normal operation of dustproof facilities that play an important role in industrial sites in real time. Currently, a fault detection technology based on a limit check is widely applied throughout the industry in order to monitor whether a rotating machine such as a scrubber is operating normally.

In the case of this limit checking technique, it is important to set a threshold value that distinguishes between failure and normality in order to prevent non-detection or false detection.

Since the threshold value varies according to the operating conditions of the device (accumulated driving time, driving speed, etc.), it is important to properly adjust the threshold value to improve the accuracy of the fault detection technology.

The background technology of the present invention is disclosed in Republic of Korea Patent Registration No. 10-2248063 (Announcement on May 3, 2021).

A technical problem to be achieved by the present invention is to provide a scrubber fan monitoring apparatus and method using an adaptive Kalman filter for extracting characteristic values in the time domain using the adaptive Kalman filter and diagnosing whether or not a scrubber fan has failed.

In addition, to provide a scrubber fan monitoring device and method using an adaptive Kalman filter to simultaneously solve the problem of extracting feature values in the time domain and adjusting the threshold required for fault detection of the scrubber fan by convergence of the Kalman filter and probability statistics theory. It is for

An apparatus for monitoring a scrubber fan using an adaptive Kalman filter according to an embodiment of the present invention to achieve this technical problem includes a storage unit for storing raw data collected from a sensor; an observation data generating unit generating observation data using the stored raw data; a Kalman filtering unit inputting the observed data to an adaptive Kalman filter and outputting estimated data for estimating the observed data from which high-frequency noise has been removed; a feature value extractor extracting normalized statistical data as feature values using the estimation data estimated by the adaptive Kalman filter; and a failure determination unit counting the number of times the extracted characteristic value exceeds a preset threshold value for a set time period to determine whether the scrubber fan is out of order.

In addition, the adaptive Kalman filter may be a recursive algorithm in which an adaptation step of estimating process noise variance is combined with a standard Kalman filter including a prediction step of predicting a state variable and a correction step of correcting an error between a measurement and a predicted value. .

In addition, the Kalman filtering unit removes high-frequency noise included in the observation data by setting the observation noise variance to a value larger than the noise variance of the actual observation data in the correction step, adapts to changes in statistical characteristics of the observation data, and In order to output the estimated data including high-frequency components, process noise variance may be estimated by the following equation in the adaptation step.

Here, Q(k) is the process noise variance, S(k) is the scaling factor, and M(k) is the residual variance.

In addition, the feature value extractor extracts normalized statistical data based on the probability statistics theory as the feature value using the estimated data estimated by the adaptive Kalman filter and related numerical data, feature values can be extracted.

는 특징값,

는 상태 교정 성분,

는 상태교정 성분의 분산,

는 칼만 이득,

는 이노베이션(innovation) 분산이다.here,

is the feature value,

is the state correction component,

is the variance of the state correction component,

is the Kalman gain,

is the innovation variance.

In addition, the failure determination unit counts the number of times the extracted characteristic value exceeds a preset threshold value for the set time period, and if the count exceeds the number of revolutions of the scrubber fan motor for the set time period, the scrubber fan It is determined as a failure of , and the threshold value may be preset as a confidence interval according to the extracted feature value.

In addition, a scrubber fan monitoring method using an adaptive Kalman filter according to another embodiment of the present invention includes storing raw data collected from a sensor; generating observation data using the stored raw data; outputting estimation data for estimating observation data from which high-frequency noise has been removed by inputting the observation data to an adaptive Kalman filter; extracting normalized statistical data as feature values using estimated data estimated by the adaptive Kalman filter; and counting the number of times the extracted characteristic value exceeds a predetermined threshold value for a set time period to determine whether or not the scrubber fan is out of order.

In this case, the adaptive Kalman filter may be a recursive algorithm in which an adaptation step of estimating a process noise covariance is combined with a standard Kalman filter including a prediction step of predicting a state variable and a correction step of correcting an error between a measurement and a predicted value. .

In addition, in the step of outputting the estimated data, the high-frequency noise included in the observation data is removed by setting the observation noise variance to a value larger than the noise variance of the actual observation data in the correction step, and adapts to changes in statistical characteristics of the observation data. In order to output the estimated data including the high-frequency component according to the defect, the process noise variance may be estimated by the following equation in the adaptation step.

Here, Q(k) is the process noise variance, S(k) is the scaling factor, and M(k) is the residual variance.

In addition, the step of extracting as the feature value extracts normalized statistical data based on the probability statistics theory as the feature value using the estimated data estimated by the adaptive Kalman filter and related numerical data, The feature value can be extracted by Eq.

는 특징값,

는 상태 교정 성분,

는 상태교정 성분의 분산,

는 칼만 이득,

는 이노베이션(innovation) 분산이다.here,

is the feature value,

is the state correction component,

is the variance of the state correction component,

is the Kalman gain,

is the innovation variance.

In addition, the step of determining whether the scrubber fan is out of order is to count the number of times the extracted feature value exceeds a preset threshold value during the set time period, and the count number is the number of rotations of the scrubber fan motor during the set time period If it exceeds , it is determined that the scrubber fan is out of order, and the threshold value may be preset as a confidence interval according to the extracted feature value.

As such, according to the present invention, it is possible to extract feature values in the time domain using the adaptive Kalman filter and diagnose whether or not the scrubber fan has a failure, thereby improving the accuracy of failure detection.

In addition, according to the present invention, the problem of extracting feature values in the time domain and adjusting the threshold required for fault detection of the scrubber fan can be solved at the same time by convergence of the Kalman filter and probability statistics theory, thereby improving the fault detection accuracy.

In addition, according to the present invention, since data is processed in real time in real time using a time domain analysis technique, monitoring can be performed with almost no processing delay compared to frequency analysis techniques, and abnormal conditions of the scrubber fan can be detected at an early stage. Therefore, it has the effect of being useful in establishing a maintenance schedule.

In addition, according to the present invention, it can be usefully used for failure detection of various rotating machine systems.

1 is a block diagram showing a scrubber fan monitoring apparatus using an adaptive Kalman filter according to an embodiment of the present invention.
FIG. 2 is a diagram showing a raw data sequence of FIG. 1 .
FIG. 3 is a diagram showing the structure of the Kalman filtering unit of FIG. 1 by way of example.
4 is a flow chart illustrating an operation flow of a scrubber fan monitoring method using an adaptive Kalman filter according to an embodiment of the present invention.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings. In this process, the thickness of lines or the size of components shown in the drawings may be exaggerated for clarity and convenience of description.

In addition, terms to be described later are terms defined in consideration of functions in the present invention, which may vary according to the intention or custom of a user or operator. Therefore, definitions of these terms will have to be made based on the content throughout this specification.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the drawings.

First, an apparatus for monitoring a scrubber fan using an adaptive Kalman filter according to an embodiment of the present invention will be described with reference to FIGS. 1 to 3 .

1 is a block diagram showing a scrubber fan monitoring apparatus using an adaptive Kalman filter according to an embodiment of the present invention.

As shown in FIG. 1, the apparatus 100 for monitoring a scrubber fan using an adaptive Kalman filter according to an embodiment of the present invention includes a storage unit 110, an observation data generation unit 120, a Kalman filtering unit 130, and a characteristic value. It includes an extraction unit 140 and a failure determination unit 150.

First, the storage unit 110 stores a certain amount of raw data collected from the sensor 200 .

At this time, the sensor 200 is attached to the scrubber fan or for measuring the noise level near the scrubber fan, and in detail, a vibration acceleration sensor for measuring the vibration acceleration of the scrubber fan and the motor side can be applied.

Also, the storage unit 110 may be a buffer.

The observation data generation unit 120 generates observation data using raw data stored in the storage unit 110 .

That is, the observation data generation unit 120 pre-processes the raw data stored in the buffer to generate observation data input to the adaptive Kalman filter, such as average, RMS, and maximum values. At this time, the statistical characteristics of the observation data may vary according to the operating time and operating speed of the system.

FIG. 2 is a diagram showing a raw data sequence of FIG. 1 .

As shown in FIG. 2, the buffer according to the embodiment of the present invention is applied with a buffer of a sliding window concept, and the raw data is filled by allocating memory on the board in the order in which raw data is received.

At this time, the observation data generating unit 120 generates the first observation data when all the data is filled in each index of the set buffer size (n), and then fills the data from the first index of the buffer again, but half of the buffer is new. When it is filled with data, the second observation data is created using all the data in the buffer, and when the other half of the buffer is subsequently updated with new data, the third observation data is created using all the data in the buffer. Thereafter, the Nth observation data is generated in the same way. Therefore, it is preferable to set the buffer size to an even number.

Therefore, there is an overlap corresponding to 50% of the buffer size (n) between the raw data used when two consecutive observation data are generated, and the reason for considering the common raw data is smoother. (smooth) This is to obtain the observation data signal waveform.

In addition, the Kalman filtering unit 130 inputs the observation data generated by the observation data generator 120 to an adaptive Kalman filter that self-updates the process noise variance, thereby estimating the observation data from which high-frequency noise has been removed (that is, estimated data). As an output of the data filtering process, estimated observational data) is output.

In the Kalman filtering unit 130 according to an embodiment of the present invention, the adaptive Kalman filter includes a prediction phase of predicting a state variable and a correction phase of correcting an error between a measurement and a predicted value. It is a recursive algorithm that combines a standard Kalman filter including ) with an adaptation phase for estimating process noise variance. In this case, the state variable to be estimated by the adaptive Kalman filter is observation data, and the model used to predict the state transition of the state variable may be a random walking model.

In detail, the standard Kalman filter consists of a prediction step and a correction step, and since the standard Kalman filter requires a priori knowledge of process noise and observation noise (assuming that the process noise variance and the observed noise variance are invariant), the performance of the Kalman filter is affected by the set value of variance.

In general, the Kalman filtering unit 130 tends to be insensitive to high-frequency fluctuations of the observation data when the observation noise variance is set high in the standard Kalman filter, and therefore, the observation noise variance of the Kalman filter is set to a value larger than the noise variance of the actual observation data. By setting it, high-frequency noise included in the observation data can be properly removed.

Meanwhile, when the observation noise variance is set to a constant value in the standard Kalman filter, the Kalman filtering unit 130 tends to be sensitive to high-frequency fluctuations of the observation data as the process noise variance increases, so that the observation noise can be properly removed. For this purpose, it is desirable to keep the variance of process noise small, and at the moment when observation data including high-frequency components due to defects is input, it is necessary to increase the variance of process noise so that the observation data including high-frequency components due to the failure can be appropriately estimated. there is.

The adaptive Kalman filter is a combination of a standard Kalman filter (prediction step + correction step) and an adaptation step for estimating process noise variance.

In the adaptive Kalman filter according to an embodiment of the present invention, the observation noise variance is assumed to be time-invariant, and the adaptive Kalman filter estimates and updates the process noise variance in real time through an adaptation step, thereby removing high-frequency noise, i.e., estimation. estimate the data.

In general, defects in high-speed rotating machines such as scrubbers cause high-frequency components in observed data. The adaptive Kalman filter according to an embodiment of the present invention appropriately removes high-frequency noise while providing observation data with periodic high-frequency components due to defects. Output after filtering (estimating).

The state variables (observed data from which high-frequency noise is removed) considered by the adaptive Kalman filter are scalar variables and the state transition model of the state variable to be estimated is not known, so in the embodiment of the present invention, the state variable We introduce a random walking model as a state transition model of .

At this time, the state variable can be expressed as in Equation 1 below.

는 상태변수,

는 평균이 0이고 분산이

인 공정 잡음이다.here,

is the state variable,

has a mean of 0 and a variance of

is process noise.

In addition, the observation model can be expressed as Equation 2 below.

는 관측데이터 생성부(120)에서 생성된 관측데이터,

는 평균이 0이고, 분산이

인 관측 잡음이다.here,

Is the observation data generated by the observation data generating unit 120,

has a mean of 0 and a variance of

is the observation noise.

Parameters shown in the recursive algorithm of the adaptive Kalman filter from Equations 1 and 2 are determined as F (transition matrix) = 1 and H (observation matrix) = 1.

The recursive algorithm of such an adaptive Kalman filter is as follows.

FIG. 3 is a diagram showing the structure of the Kalman filtering unit 130 of FIG. 1 by way of example.

As shown in FIG. 3, the adaptive Kalman filter is performed in the order of a prediction step, a correction step, and an adaptation step. First, in the prediction step, a state variable and an estimated variance can be predicted as shown in Equation 3 below.

는 무작위 보행 모델에 의해 스텝 k-1에서 추정한 상태변수가 스텝 k에서 상태변수의 예측값이다. 그리고

는 스텝 k-1에서 추정된 P(k-1)과 공정 잡음 분산

의 합으로 스텝 k의 추정 분산 예측값이다.At this time,

is the state variable estimated at step k-1 by the random walk model and the predicted value of the state variable at step k. and

is the estimated P(k-1) at step k-1 and the process noise variance

The sum of is the estimated variance prediction value of step k.

And in the correction step, the following values can be calculated by Equation 4, respectively.

는 관측데이터와 예측값 사이의 오차(즉, 이노베이션(innovation)이고,

는 예측한 추정 분산과 관측 잡음 분산과의 합(즉, 이노베이션 분산),

는 예측한 추정 분산과 이노베이션 분산의 비(즉, 칼만 이득(Kalman Gain)),

는 이노베이션과 칼만 이득으로 예측값을 보정하여 계산되는 상태변수 추정값,

는 칼만 이득을 이용하여 예측한 추정 분산을 보정한 추정 분산이다.At this time,

is the error between the observed data and the predicted value (i.e., innovation),

is the sum of the predicted estimated variance and the observed noise variance (i.e., the innovation variance),

is the ratio of the predicted estimated variance and the innovation variance (i.e., Kalman Gain),

is the state variable estimate calculated by correcting the predicted value with the innovation and Kalman gain,

is the estimated variance corrected for the estimated variance predicted using the Kalman gain.

In Fig. 3, the observed noise covariance is assumed to be constant at every step k (i.e., R(k)=R).

).In addition, in general, the Kalman filtering technique tends to trust the observed data more than the predicted value as the observation noise variance (R) is small. For example, if the observation noise variance is set to R = 0, the Kalman gain K (k) = 1, and the Kalman filter outputs observation data containing high-frequency noise as an estimated value (ie,

).

Therefore, in an embodiment of the present invention, in order for the adaptive Kalman filter to filter out the high-frequency noise included in the observation data, the observation noise variance R is set to a value relatively larger than the noise variance of the actual observation data derived from the sensor 200. At this time, if the observation noise variance (R) is set too large, sensitivity to high-frequency signals decreases, so it is desirable to set it appropriately to suit the type of device and sensor. That is, the above recommendation recalls that in an embodiment of the present invention, the sensitivity of the Kalman filter to high-frequency noise and signals can be preceded before the algorithm is executed by adjusting the observed noise variance (R).

In addition, as the operating conditions of the system change, the statistical characteristics (variance) of the observed data may change from moment to moment. At this time, the factor of change in the variance of the observed data should not be interpreted as a change in the statistical characteristics (variance) of the noise included in the observed data. As described above, the observed noise variance is always constant in the adaptive Kalman filter. Therefore, in an embodiment of the present invention, the adaptive Kalman filter considers the change in the variance of the observation data as a change in the model uncertainty (ie, process noise variance) in the state transition model (ie, the random walk model), and the state variable (observed Data) It is preferable to include an adaptation step of estimating the process noise variance to vary the process noise variance required for estimation.

Finally, in the adaptation step, process noise variance can be estimated by Equation 5 below.

Here, Q(k) is the process noise variance, S(k) is the scaling factor, and M(k) is the residual variance.

In this case, the residual means an error between the estimated value of the state variable of the Kalman filter and the observed data.

Process noise variance and residual variance are positively correlated. Accordingly, one of the methods for estimating the process noise variance may be to use the residual variance estimated at step k as the process noise variance at k+1. Using residual data within a sliding observation window of size N, the residual covariance is estimated as shown in Equation 6 below.

이고, 적응형 칼만필터의 초기화 단계에서 j<0에 대한

는 모두 0으로 설정하는 것이 바람직하다.here,

, and for j<0 in the initialization step of the adaptive Kalman filter

It is desirable to set all to 0.

Since R is fixed at the initialization stage in consideration of the noise level of the actual observed data, in a steady-state condition, the Kalman filter becomes insensitive to high-frequency fluctuations of the observed data as the process noise variance decreases, the Kalman gain decreases, In the opposite case, as the Kalman gain increases, it becomes sensitive to high-frequency fluctuations in the observation data. In order for the adaptive Kalman filter to respond sensitively to periodic high-frequency signals caused by defects while being insensitive to high-frequency observation noise, the variable rate of process noise variance is important. The rate of change of the process noise covariance is related to the convergence of the Kalman filter.

Therefore, the convergence speed of the adaptive Kalman filter is increased by additionally compensating the residual covariance using a scaling (fading) factor so that the process noise covariance can increase and decrease rapidly (S(k) as M(k) as shown in Equation 5). by multiplying by ) can be improved.

)이 평균이 0이고 공분산이 G(k)인 확률 변수' 라는 가정을 기반으로 유도되며 다음의 수학식 7과 같이 결정된다.At this time, the scaling factor is the 'state correction component (

) is derived based on the assumption that 'a random variable with a mean of 0 and a covariance of G(k)' is determined by Equation 7 below.

와

는 각각 상태교정 성분의 분산(G(k))을 관측 창 내의 데이터를 이용하여 수치적으로 계산된 추정값과 정상상태 가정을 기반으로 칼만필터가 가진 통계적 정보를 이용하여 얻은 근사값으로, 각각 다음의 수학식 8과 같이 구할 수 있다.here,

and

is an approximate value obtained by using the statistical information possessed by the Kalman filter based on the estimated value calculated numerically using the data within the observation window and the steady-state assumption of the variance (G(k)) of each state correction component, respectively. It can be obtained as in Equation 8.

는 정상상태에서 칼만 이득이 일정(constant)하다는 가정과 함께 관계식

와 정의식

을 사용하여 다음의 수학식 9와 같이 유도될 수 있다.At this time,

is the relational expression with the assumption that the Kalman gain is constant in the steady state

and Jung Eui-sik

It can be derived as in Equation 9 below using

The feature value extraction unit 140 extracts normalized statistical data as feature values using the estimated data estimated by the Kalman filtering unit 130 and related numerical data (by-products).

At this time, the feature value extractor 140 extracts the estimated data estimated by the adaptive Kalman filter and related numerical data (by-product data generated in the process of outputting estimated data by the adaptive Kalman filter, such as predicted values, Kalman gains, and innovation variances). Normalized statistical data based on the probability statistics theory are extracted as feature values using Equation 10 below.

는 특징값,

는 상태 교정 성분,

는 상태교정 성분의 분산,

는 칼만 이득,

는 이노베이션(innovation) 분산이다. here,

is the feature value,

is the state correction component,

is the variance of the state correction component,

is the Kalman gain,

is the innovation variance.

시퀸스가 표준정규분포를 따른다면

는 1 자유도를 가지는 카이자승(Chi-square) 분포를 따른다. 1 자유도 카이자승 분포의 99.5% 신뢰구간을 고려할 때, upper-tail 임계값은 7.879이다.if,

If the sequence follows the standard normal distribution

follows a Chi-square distribution with one degree of freedom. Considering the 99.5% confidence interval of the chi-square distribution with 1 degree of freedom, the upper-tail critical value is 7.879.

는 99.5%의 정상 데이터이며,

는 0.5%의 비정상 데이터로 구분할 수 있다.thus,

is 99.5% normal data,

can be classified as 0.5% of abnormal data.

Finally, the failure determination unit 150 counts the number of times that the feature value extracted by the feature value extraction unit 140 exceeds a preset threshold value for a set time period, and determines whether the scrubber fan is out of order.

At this time, the set time is the time required for fault diagnosis and is a time preset as a monitoring cycle.

That is, the failure determination unit 150 counts the number of times the feature value extracted by the feature value extraction unit 140 exceeds a preset threshold value for a set time period, and the counted number is the number of rotations of the scrubber fan motor during the set time period. If it exceeds , it is judged to be a failure of the scrubber fan.

In other words, the failure determination unit 150 determines whether the scrubber fan is out of order by detecting outliers in which the feature value extracted by the feature value extraction unit 140 deviate from a fixed threshold of the concept of a confidence interval.

At this time, considering that the scrubber fan motor is operated within ±10% of the rated rpm for safety reasons, the rotation speed of the scrubber fan motor is determined during the time required for fault diagnosis using the motor RPM data sensed by the sensor 200. The number of revolutions of the motor can be found.

In addition, the threshold value may be preset as a confidence interval according to the feature value extracted as described above.

Hereinafter, a scrubber fan monitoring method using an adaptive Kalman filter according to an embodiment of the present invention will be described with reference to FIG. 4 .

FIG. 4 is a flow chart illustrating an operational flow of a method for monitoring a scrubber fan using an adaptive Kalman filter according to an embodiment of the present invention. Referring to this flow chart, specific operations of the present invention will be described.

According to an embodiment of the present invention, first, the storage unit 110 stores raw data collected from the sensor 200 (S10).

Next, the observation data generation unit 120 generates observation data using the raw data stored in step S10 (S20).

Next, the observation data generated in step S20 is input to the Kalman filtering unit 130, and the adaptive Kalman filter outputs estimated data for estimating the observation data from which high-frequency noise has been removed (S30).

An adaptive Kalman filter according to an embodiment of the present invention is a recursion combining a standard Kalman filter including a prediction step of predicting state variables and a correction step of correcting errors between measured and predicted values, and an adaptation step of estimating the covariance of process noise. It is an algorithm. In this case, the state variable to be estimated by the adaptive Kalman filter may be a scalar variable or observed data, and a model used to predict the state transition of the state variable may be a random walking model.

Therefore, in step S30, the Kalman filtering unit 130 sets the observation noise variance to a larger value than the noise variance of the actual observation data in the correction step to properly remove the high-frequency noise included in the observation data, and in the prediction step (step k+ By estimating and using the process noise variance required in 1) in real time in the adaptation step (step k), it is possible to adapt to changes in the statistical characteristics of the observation data and appropriately estimate the observation data including high-frequency components according to defects. The process noise variance is estimated by Equation 5 in the adaptation step.

Next, the feature value extraction unit 140 extracts normalized statistical data as feature values using the estimated data and numerical data estimated in step S30 (S40).

In detail, in step S40, the feature value extraction unit 140 extracts normalized statistical data based on the probability statistics theory as feature values using the estimated data and numerical data, and calculates the feature values by Equation 10 above. extract

Next, the failure determination unit 150 counts the number of times the feature value extracted in step S40 exceeds a preset threshold value during the set time (S50 and S60).

At this time, the failure determination unit 150 determines whether the number of times counted during the set time in step S60 exceeds the number of revolutions of the scrubber fan motor (S70).

As a result of the determination in step S70, if the number of counts in step S60 exceeds the number of rotations of the scrubber fan motor for the set time, the failure determining unit 150 determines that the failure of the scrubber fan has occurred (S80).

If the number of times counted during the set time in step S60 does not exceed the number of revolutions of the scrubber fan motor, the failure determination unit 150 determines that the scrubber fan is normally driven (S90).

In this case, the threshold value may be preset as a confidence interval according to the feature value extracted as described above.

In an embodiment of the present invention, the adaptive Kalman filter solves the problem of setting a threshold value that is not clear and can be varied according to the operating conditions of the system in determining whether there is a failure. In detail, by using the estimated data and numerical data output from the adaptive Kalman filter, a feature value following a chi-square distribution is extracted through a normalization process, and the reliability of the feature value is determined by setting a fixed confidence interval as a threshold value. The threshold variable problem can be solved.

Such a scrubber fan monitoring apparatus and method using an adaptive Kalman filter may be implemented as an application or implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer readable recording medium may include program instructions, data files, data structures, etc. alone or in combination.

Program instructions recorded on the computer-readable recording medium may be those specially designed and configured for the present invention, or those known and usable to those skilled in the art of computer software.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floptical disks. media), and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like.

Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter or the like as well as machine language codes generated by a compiler. The hardware device may be configured to act as one or more software modules to perform processing according to the present invention.

As described above, the scrubber fan monitoring apparatus and method using the adaptive Kalman filter according to the embodiment of the present invention can extract feature values in the time domain using the adaptive Kalman filter and diagnose whether the scrubber fan is out of order. This can improve the accuracy of fault detection.

In addition, according to an embodiment of the present invention, the problem of extracting feature values in the time domain and adjusting the threshold required for fault detection of the scrubber fan can be simultaneously solved by convergence of the Kalman filter and probability statistics theory, thereby improving fault detection accuracy.

In addition, according to the embodiment of the present invention, since data is processed in real time in real time using the time domain analysis method, monitoring can be performed with little processing delay compared to the frequency analysis method, and abnormal conditions of the scrubber fan can be detected early. Therefore, it can be usefully used for planning maintenance schedules.

In addition, according to an embodiment of the present invention, it can be usefully used for failure detection of various rotating machine systems.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is merely exemplary, and those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. will be. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the claims below.

100: scrubber fan monitoring device
110: storage unit 120: observation data generation unit
130: Kalman filtering unit 140: feature value extraction unit
150: failure determination unit 200: sensor

Claims (10)

In the scrubber fan monitoring device using an adaptive Kalman filter,
a storage unit for storing raw data collected from the sensor;
an observation data generating unit generating observation data using the stored raw data;
a Kalman filtering unit inputting the observed data to an adaptive Kalman filter and outputting estimated data for estimating the observed data from which high-frequency noise has been removed;
a feature value extractor extracting normalized statistical data as feature values using the estimation data estimated by the adaptive Kalman filter; and
Scrubber fan monitoring device comprising a failure determination unit for counting the number of times the extracted characteristic value exceeds a preset threshold value for a set time period to determine whether or not the scrubber fan is out of order. According to claim 1,
The adaptive Kalman filter,
A scrubber fan monitoring device, which is a recursive algorithm combining a standard Kalman filter including a prediction step of predicting state variables and a correction step of correcting errors between measured and predicted values, and an adaptation step of estimating process noise variance. According to claim 2,
The Kalman filtering unit,
In the correction step, the high-frequency noise included in the observed data is removed by setting the observation noise variance to a value larger than the noise variance of the actual observation data, adapting to changes in the statistical characteristics of the observation data, and the high-frequency component included in the defect. A scrubber fan monitoring device for estimating the process noise variance by the following equation in the adaptation step to output estimated data:

Here, Q(k) is the process noise variance, S(k) is the scaling factor, and M(k) is the residual variance. According to claim 1,
The feature value extraction unit,
Extracting normalized statistical data based on probability statistical theory as the feature value using the estimated data estimated by the adaptive Kalman filter and related numerical data,
A scrubber fan monitoring device for extracting the feature value by the following equation:

here,

is the feature value,

is the state correction component,

is the variance of the state correction component,

is the Kalman gain,

is the innovation variance. According to claim 1,
The failure determination unit,
Counting the number of times the extracted feature value exceeds a preset threshold value during the set time period, and determining that the scrubber fan is out of order when the count exceeds the number of revolutions of the scrubber fan motor during the set time period;
The threshold is
A scrubber fan monitoring device that is preset as a confidence interval according to the extracted feature value. In the scrubber fan monitoring method performed by a scrubber fan monitoring device using an adaptive Kalman filter,
Storing the raw data collected from the sensor;
generating observation data using the stored raw data;
outputting estimation data for estimating observation data from which high-frequency noise has been removed by inputting the observation data to an adaptive Kalman filter;
extracting normalized statistical data as feature values using estimated data estimated by the adaptive Kalman filter; and
A scrubber fan monitoring method comprising counting the number of times the extracted characteristic value exceeds a preset threshold value for a set time period to determine whether the scrubber fan is out of order. According to claim 6,
The adaptive Kalman filter,
A scrubber fan monitoring method, which is a recursive algorithm combining a standard Kalman filter including a prediction step of predicting state variables and a correction step of correcting errors between measured and predicted values, and an adaptation step of estimating process noise variance. According to claim 7,
outputting the estimated data,
In the correction step, the high-frequency noise included in the observed data is removed by setting the observation noise variance to a value larger than the noise variance of the actual observation data, adapting to changes in the statistical characteristics of the observation data, and the high-frequency component included in the defect. A scrubber fan monitoring method for estimating process noise variance by the following equation in the adaptation step to output estimated data:

Here, Q(k) is the process noise variance, S(k) is the scaling factor, and M(k) is the residual variance. According to claim 6,
The step of extracting the feature value,
Extracting normalized statistical data based on probability statistical theory as the feature value using the estimated data estimated by the adaptive Kalman filter and related numerical data,
Scrubber fan monitoring method for extracting the feature value by the following equation:

here,

is the feature value,

is the state correction component,

is the variance of the state correction component,

is the Kalman gain,

is the innovation variance. According to claim 6,
The step of determining whether the scrubber fan is out of order,
Counting the number of times the extracted feature value exceeds a preset threshold value during the set time period, and determining that the scrubber fan is out of order when the count exceeds the number of revolutions of the scrubber fan motor during the set time period;
The threshold is
A scrubber fan monitoring method that is preset as a confidence interval according to the extracted feature value.

Images