KR102093929B1 - Apparatus and Method for Diagnosing Mechanical System Health based on CIM - Google Patents

Abstract

The present invention relates to an apparatus for diagnosing the health of a system based on a critical information map (CIM), which can use the CIM to find out an area including a signal affecting a malfunction in time-frequency representations (TFRs), and method thereof. According to the present invention, the apparatus for diagnosing the health of the system based on the CIM comprises: a data synchronization unit which extracts only the signals of a relevant frequency representing a non-stationary movement by a band-pass filter, and uses the cross-correlation analysis as a way to identify the degree to which two types of serial signals are equal to each other; a subtracted spectrogram unit which uses the wavelet packet decomposition (WPD) as a TFRs method for expressing the CIM, makes the signals into a spectrum, and performs an inversion after subtracting; and a map value optimization unit which determines parameters which determine the positions of important areas and the size of each area in the subtracted spectrogram to make the CIM.

Description

Apparatus and Method for Diagnosing Mechanical System Health based on CIM

The present invention relates to system health diagnosis, specifically, CIM-based system health diagnosis that enables to find a region including a signal affecting a failure in TFRs (Time Frequency Representations) using a CIM (Critical Information Map). It relates to a device and method for.

In the field of engineering, the resolution of uncertainties that can cause serious damage or lead to serious accidents has been regarded as an ongoing task.

In particular, in the manufacturing industry, a sudden failure of one production equipment can lead to the suspension of the entire production line. In the manufacturing industry, there is a continuing demand for predicting the failure or longevity of equipment having such uncertainty.

With such industrial demands, with the development of data science, computer performance has improved, and research into a diagnosis system for prognosis of failure of mechanical equipment has been actively conducted.

In general, in order to diagnose a machine, it is necessary to use various signals such as a signal indicating the state of the machine and a sensor to measure the temperature, voltage, pressure, and acceleration.

Among them, a voltage sensor and an acceleration sensor are frequently used. Unlike data with small changes such as temperature and pressure, data can be received at a high sampling frequency due to its instantaneous change and dynamic characteristics.

This means that a lot of information can be obtained only with short-term measurement data, which can be used as very useful data for diagnosing machines.

In the diagnostic engineering using the vibration of the prior art, data of the gearbox and the bearing were mainly used.

Mitchell Lebold wrote a signal processing method to extract feature elements for diagnosing conditions from gearbox vibrations.

As a first step, in the raw signal state, statistical data such as rms, kurtosis, and skweness are extracted as important information of the failure and data are utilized.

In the second step, a representative value is extracted from a signal subjected to the processing of Time Synchronous Averaging (TSA), which is a method of removing noise of a repeated signal using an encoder or tachometer that can know the speed of the gear.

Subsequently, the gear mesh frequency is removed to make a residual signal (RES), or the side band of the gear mesh frequency is additionally removed to extract the band frequency from the remaining signal, DIF (Difference Signal) or TSA signal, and a band-pass mesh signal. ).

In addition, Wei He at al incorporated the related frequency density index for each part of the bearing called bearing characteristic frequencies and utilized it as a characteristic factor of failure, and also combined statistical values such as kurtosis to grasp the degree of bearing degradation and defects.

However, these conventional diagnostic methods using vibration are methods that can be used only by knowing the operating frequency and the characteristics of the bearing or the gear directly, and are applicable only to signals in a stationary state.

In order to access the vibration signal as data itself, a bearing diagnosis method may be used using Time-Frequency Representations (TFRs) and Convolutional Neural Network (CNN).

David Verstraete transformed a one-dimensional signal into a two-dimensional image using three methods: Short Time Fourier Transform (STFT), Wavelet Transform (WT) and Hilbert-Huang Transform (HHT), and used it as an input to CNN.

This is because the accuracy of diagnosis was high, but because the data was not processed directly, there was a lot of noise. Therefore, a large amount of data is required to learn directly, and the variation in results is very large according to the 2D imaging method. there is a problem.

Accordingly, there is a need to develop a new system health diagnosis technology that outputs robust diagnosis results even without knowing the operating frequency and characteristics of the diagnosis object and does not require a huge number of data for learning.

Republic of Korea Patent Publication No. 10-2012-0027733 Republic of Korea Registered Patent No. 10-1129466

The present invention is to solve the problems of the system health diagnosis technology of the prior art, by using a CIM (Critical Information Map) to find the region containing the signal affecting the failure in the TFRs (Time Frequency Representations) The aim is to provide an apparatus and method for diagnosing system health based on CIM.

An object of the present invention is to provide an apparatus and method for system health diagnosis based on CIM that enables robust diagnostic results to be output even without knowing the internal information of the equipment in detail by using a data-based approach. .

An object of the present invention is to provide an apparatus and method for system health diagnosis based on CIM that can effectively output a diagnosis result even in a non-stationary signal by conserving and using a domain of TFRs.

An object of the present invention is to provide an apparatus and method for system health diagnosis based on CIM that enables system health diagnosis even when learning is performed using fewer data than the CNN-TFRs method.

Other objects of the present invention are not limited to those mentioned above, and other objects not mentioned will be clearly understood by those skilled in the art from the following description.

In order to achieve the above object, the apparatus for diagnosing system health based on CIM according to the present invention extracts only signals of a corresponding frequency representing a non-stationary movement with a band pass filter, and two types of time series signals. A data synchronization unit that uses Cross Correlation as a method of grasping the degree of agreement of the data; using Wavelet Packet Decomposition (WPD) as a Time Frequency Representations (TPRs) method for expressing a Critical Information Map (CIM), and Subtracted spectrogram unit that uses spectral subtraction to reverse transform after spectralization and subtraction; to create a critical information map (CIM), determine the location of the important region and the size of each region in the subtracted spectrogram. Characterized in that it comprises; a map value optimization unit for determining the parameters to determine.

A method for diagnosing system health based on CIM according to the present invention for achieving another object extracts only signals of a corresponding frequency representing a non-stationary movement with a band pass filter and matches two types of time series signals. Data Synchronization step using Cross Correlation as a method to grasp the degree; Wavelet Packet Decomposition (WPD) is used as a Time Frequency Representations (TPRs) method for representing a Critical Information Map (CIM), Generating a subtracted spectrogram using spectral subtraction, which inversely transforms the signal after spectralizing the signal; generating a critical information map in the subtracted spectrogram to create a critical information map (CIM) Map value optimization to determine parameters that determine the location and size of each area (Map V alue Optimization) step; characterized in that it comprises a.

An apparatus and method for diagnosing system health based on CIM according to the present invention as described above has the following effects.

First, by using a CIM (Critical Information Map), it is possible to efficiently find a region including a signal affecting a failure in Time Frequency Representations (TFRs).

Second, by using a data-based approach, it is possible to output robust diagnostic results even without knowing the internal information of the equipment in detail.

Third, by preserving and using the domain of TFRs, it is possible to effectively output a diagnosis result even in a non-stationary signal.

Fourth, system health diagnosis is possible even when learning is performed using fewer data than the CNN-TFRs method.

1 is a block diagram of an apparatus for diagnosing system health based on CIM according to the present invention;
Figure 2 shows the spectral form (a) STFF, (b) WT, (c) WPD showing TFRs in the signal.
3 is a block diagram showing an example of a subtracted spectrogram
4 is a configuration diagram showing an example of the second and third parameter settings
5 is a block diagram showing the position of the input gear and the accelerometer of the second axis of the articulated robot
6 is a configuration diagram showing a robot data acquisition motion
7A to 7C are acquired raw data and raw data spectrograms and filtered data graphs.
8 is a time synchronization graph of low signal data
9A to 9C are reference spectrograms and NR spectrograms and AR spectrograms.
10 is a flow chart showing a data classification process according to the fitting degree
FIG. 11A shows the learned CIM results (NR), FIG. 11B shows the learned CIM results (SR and HR), FIG. 11C shows the CIM windows (NR and AR), and FIG. 11D shows the CIM windows (SR and HR). Structure shown
FIG. 12A is a SS coefficient (NR window) of the training data set, and FIG. 12B is a SS coefficient (AR window) graph of the training data set.

Hereinafter, a preferred embodiment of an apparatus and method for diagnosing system health based on CIM according to the present invention will be described in detail as follows.

The features and advantages of the apparatus and method for CIM-based system health diagnosis according to the present invention will become apparent through detailed description of each embodiment below.

1 is a configuration diagram of a device for diagnosing system health based on CIM according to the present invention.

An apparatus and method for diagnosing system health based on CIM according to the present invention is such that a region including a signal affecting a failure can be found in time frequency representations (TFRs) using a CIM (Critical Information Map).

The present invention can provide a signal processing method applicable to a non-stationary signal using less data even without knowing internal information by using a CIM (Critical Information Map).

The present invention is to find a region containing a signal that significantly affects the failure in the TFRs by using a CIM (Critical Information Map),

(i) Data Synchronization using Band Pass Filter extraction and Cross Correlation

(ii) Subtracted Spectrogram using Wavelet Packet Decomposition (WPD) and Spectral Subtraction

(iii) In the subtracted spectrogram, it includes the configuration of Map Value Optimization, which determines the parameters of the location of important areas and the size of each area.

The apparatus for diagnosing system health based on CIM according to the present invention, as shown in FIG. 1, extracts only signals of a corresponding frequency representing a non-stationary movement with a band pass filter, and matches two types of time series signals. Data Synchronization (100) using Cross Correlation as a method of grasping the degree, and Wavelet Packet Decomposition (WPD) as a Time Frequency Representations (TPRs) method for representing a Critical Information Map (CIM) Subtracted Spectrogram (200) using Spectral Subtraction, which transforms the signal after spectralizing and subtracting the signal, and a Subtracted Spectrogram (CIM) to create a Critical Information Map (CIM) Subtracted Spectrogram: Map Value Optimi, which determines parameters that determine the location of important areas and the size of each area zation) unit 300.

First, the data synchronization (Data Synchronization) process in detail as follows.

The randomly collected data has a time delay, which may cause subtraction between data rather than the same operating state.

In order to solve this, it is necessary to synchronize the data, and then cross-correlation is used to synchronize the data.

First, in frequency filtering, signals that are not related to non-stationary movement become noise and interfere with the sync when data is synchronized by using only cross correlation.

Therefore, in order to extract only signals representative of the non-stationary movement, only signals of the corresponding frequency are extracted with a band pass filter.

When processing signals, it is often the case that the sync of the signals must match.

In this case, Cross Correlation is used as a method to determine the degree of agreement between two types of time series signals.

This derives a value for the correlation between two signals at a specific sampling delay, and it is possible to match the sync of the two signals by finding the delay value that has the maximum value and applying it to the signal.

In the case of data having a non-stationary signal such as a robot, a specific motion can be found through this.

Where f is the original function g is a filtering function and τ is the time delay of the two signals. The bar indicated by the value of the definition shows the degree of agreement between the two graphs. Unlike convolution, the law of exchange does not hold.

τ _max is the time delay when the value of cross-correlation is the maximum, and the Argmax function is an expression that derives the value of the domain where the expression in parentheses becomes the maximum. Synchronize data by applying the delay of τ _max to the raw signal.

The subtracted spectrogram is as follows.

2 is a spectral form (a) STFF, (b) WT, (c) WPD showing TFRs in a signal, and FIG. 3 is a configuration diagram showing an example of subtracted spectrogram.

In order to make CIM, data conversion to TFRs, which is a two-dimensional data representation method, is required. Thereafter, SS (Spectral Subtraction) is used as a primary data processing method for extracting an area including important information.

There are several ways to represent TFRs in a signal.

Among them, the linear method is representative of a Short Time Fourier Transform (STFT), a Wavelet Transform (WT), and a Wavelet Packet Decomposition (WPD).

The main difference between these methods is the shape of the filter and the shape of the resulting tile (window) of the spectrum.

In the case of STFT, the tile size is determined in a predetermined shape in advance, and a frequency density value is derived from each tile. This occurs because the filter does not meet the required tile size, resulting in a decrease in the reliability of the relative high-frequency region and the relatively low-frequency region.

The results of David Verstraete's paper also showed that the robustness of the diagnostic algorithm was very poor when STFT was used as a conversion to a two-dimensional image for CNN learning. To compensate for this, the method of expressing the tile size differently for each frequency is WT.

However, in this method, the resolution of the time domain and the frequency domain of the tile varies depending on the number of divisions. This is not appropriate for map creation because it does not divide the window evenly.

WPD complements these problems.

It has a filter for each divided frequency. This means that the tile can be set to the desired shape in each area, and it means that a more reliable frequency density value can be obtained, unlike STFT, which is collectively converted into one filter.

Therefore, in the present invention, WPD is used as a TFRs method for CIM expression.

The kernel function of a wavelet packet that decomposes a signal into multiple frequencies in WPD is as follows.

Equation 3 includes three positive integers, j is an index scale, k is a transform operation and n is a modulation parameter or vibration parameter.

The first and second wavelet packet functions need to be predefined.

Equations 4 and 5 are referred to as a normal scaling function and a parent function, respectively.

More functions can be created like this:

Here, h (k) and g (k) are orthogonal mirror filters orthogonal to each other.

The wavelet packet coefficients in the j, n, and k states indicating the density of the filter are as follows.

Where f (t) is the time signal to be analyzed.

The value of the wavelet packet count is indicated in the WPD spectrogram of FIG. 2 (c) to be used to implement CIM.

The spectral subtraction is as follows.

In order to remove noise in acoustics, a method of improving the sound quality of a voice-existing area by extracting a signal in a voiceless area is used.

At this time, the method used is SS (Spectral Subtraction), which inverse transforms the signal after it is subtracted.

Here, P _s (w) is a noise spectrum mixed with noise, P _n (w) is a noise spectrum measured in a silent state, and P ' _s (w) is Fourier inverse transformed to obtain a noise-removed voice signal. You can.

SS is extended to WT, which is one of the TFRs that reflect time, and is expressed as follows.

는 강조된 음성(enhanced speech)의 웨이브렛 스펙트럼,

는 관찰된 신호의 웨이브렛 스펙트럼,

는 노이즈의 웨이브렛 스펙트럼이다.here,

Is the wavelet spectrum of the enhanced speech,

Is the wavelet spectrum of the observed signal,

Is the wavelet spectrum of the noise.

In addition, α is a reduction factor. And a and b are filter position and scale parameters for extracting one coefficient of the wavelet transform.

The present invention uses spectral subtraction as one means for selecting an important area in a map.

The fact that the signal from a working device is subtracted means that the specific components of the data have been removed.

However, the difference between the normally operating data was also used because it is possible to obtain error information that can appear even under the same conditions, such as random error or measurement error.

Hereinafter, a spectrogram obtained by subtracting a reference signal from normal gear data is referred to as N-R (Normal-Reference), and a reference signal subtracted from faulty gear data is referred to as A-R (Abnormal-Reference).

The description of Map Value Optimization is as follows.

3 is a block diagram showing an example of a subtracted spectrogram, and FIG. 4 is a block diagram showing an example of setting second and third variables.

In order to make a CIM, it is necessary to determine the parameters of the location of the important region and the size of each region in the subtracted spectrogram.

In this step, the process of removing areas where all data can be different due to random or measurement errors is performed in this step, and the area selected for minimum information loss is optimized by setting the objective function to be maximized.

As shown in Table 1, three variables are selected.

The first variable is the factor that determines the size of the area mentioned above.

Second, the variable 2 is selected as a threshold for the density coefficient of one subtracted spectrogram to be considered as an effective value.

Finally, the number threshold for the number of valid values in the divided rectangle is selected as the third variable.

As in FIG. 4, it is assumed that the window is determined by the first variable, and thus there are 10 frequency density values in one window.

Then, variable 2 presents the criteria for the value to be selected among the density values to select the values. The variable 3 determines whether the corresponding window is valid from the number of values selected.

The objective function should be constructed with the determined values. First, the purpose of this optimization is twofold.

The first is to select the widest possible area to minimize the loss of information. The expression for area can be expressed as follows.

y is the number of windows selected. The normalized area can be calculated by dividing this by the variable x ₁ of the number of window divisions.

The purpose of the second optimization is to eliminate areas where signals appear in the N-R as much as possible. The meaning of the signal in this area is not related to failure, but it can be said to be a variable that can occur during signal acquisition.

Figure 4 shows a simple example illustrating the definition of x ₂ and x ₃ variables.

The window in this example contains 10 parameters (ie, cells) with assigned values as shown in the figure.

Table 2 summarizes the problems in CIM optimization.

Table 2 shows the formula for the optimization problem to build the CIM.

The goal of this problem is to minimize the loss of important information.

Thus, the objective function is to maximize a normalized meaningful region (i.e., an important region), which is the number of meaningful windows normalized by the number of divisions in the time domain.

In the method for diagnosing system health based on CIM according to the present invention having such a configuration, only signals of a corresponding frequency representing a non-stationary movement are extracted with a band pass filter, and the degree of agreement between the two types of time series signals Data Synchronization step using Cross Correlation as a method of grasping; using Wavelet Packet Decomposition (WPD) as a Time Frequency Representations (TFRs) method for representing a Critical Information Map (CIM), and signaling Subtracted Spectrogram generation step using Spectral Subtraction, which is inversely transformed after spectralizing and subtracting; Positioning of important regions in the Subtracted Spectrogram to create a Critical Information Map (CIM) And map value optimization to determine parameters that determine the size of each area (Map Value O ptimization) includes;

When explaining an embodiment to which the apparatus and method for diagnosing system health based on the CIM according to the present invention described above is applied, as follows.

6 is a configuration diagram illustrating a data acquisition motion of a robot, and FIGS. 7A to 7C are graphs of the acquired raw data and raw data, and filtered data graphs.

And Figure 8 is a time synchronization graph of the raw signal data.

Apparatus and method for CIM-based system health diagnosis according to the present invention were applied to a manufacturing robot system.

The robot system is suitable for applying the present invention to show a non-stationary signal in the cycle of the motor and gear stages and a complex signal with a combination of multiple gears and belts.

In the robot used as a fault diagnosis object, a 2-axis input gear of the robot was targeted as shown in FIG. 5.

In an articulated robot, the 2nd axis is the second largest axis of the robot and is involved in the vertical motion. Two of the three input gears were damaged.

Thereafter, the accelerometer was attached to the position of FIG. 5, which is a radial direction of 2 axes, to acquire data.

As shown in Fig. 6, the data acquisition motion was reciprocated 3 times with 60 deg on 2 axes in the basic stop state.

A total of 900 sets of acceleration data were obtained by performing 300 data acquisition motions for each input gear. Acquired data at a length of 20 seconds and a sampling frequency of 12.8 kHz. The details are shown in the table below.

7A is a signal obtained from a robot equipped with a normal input gear.

The short and large signal is acquired when the robot moves down, and the long measured signal is the signal acquired when the robot moves upward.

In order to find a frequency region advantageous for synchronization in the corresponding signal, a continuous spectrum shown in FIG. 7B is drawn, and the frequency band indicated by the red square is extracted with a band-pass filter and shown in FIG. 7C.

Fig. 8 shows some of the delay calculated from the extracted signal by applying it to the raw signal.

The SS of the synchronized data and the reference signal is as follows.

9A to 9C are reference spectrograms and N-R spectrograms and A-R spectrograms.

The data that performed the corresponding process are displayed, and the area indicated in purple is a region having a high value.

The N-R signal appears over an area with a small effective area, but the A-R has an effective value over a large area. In order to remove randomly appearing areas from all the data mentioned in the previous chapter, it is necessary to remove the areas shown in FIG. 9B.

From that point of view, the optimization objective function to be performed later is weighted by the penalty from the relevant area.

FIG. 10 is a flow chart showing a data classification process according to the fitting degree, FIG. 11A is a learned CIM result (NR), FIG. 11B is a learned CIM result (SR and HR), and FIG. 11C is a CIM window (NR) And AR), FIG. 11D is a configuration diagram showing CIM windows SR and HR.

As shown in FIG. 10, the defect conditions of the gear (ie, soft and hard fitting) are classified.

First, the input device is classified as normal or abnormal, and then the abnormal gear is classified as a soft fitting or hard fitting defect.

Of the 900 data sets measured, only 5% are used to implement two CIMs, which are 15 data sets of normal soft fitting and hard fitting.

The remaining data sets (ie, 855 sets) can be used for validation.

The data set for CIM implementation includes a data synchronization step; Generating a subtracted spectrogram; Pre-processing is performed according to the Map Value Optimization step.

For CIM optimization, two parameters C _p need to be specified by the user.

As the value of these parameters increases, a meaningful window (OW) is selected in a conservative way.

Generally, C _p is set to one or more to reduce the sensitivity random error.

In this formula, set C _p = 1.5 and select OW with R _c ≥ 0.95.

Table 4 shows the optimal design for CIM.

11A and 11B are learned CIM results, and Table 4 shows design variables determined in the optimization process.

It shows that the two CIMs are colored with the meaningful R _c of each window.

If the color of the window is red (ie close to 1), the window can be a meaningful window. The window in the figure can contain a large number of SS parameters.

The CIM includes 350 and 951 SS parameters in each window of FIGS. 11A and 11B.

If the window (OW) over 0.95, which is the R _c criterion used in learning, is highlighted, it is as in FIGS. 11C and 11D.

The number of CIM OWs for N-R to A-R shown in FIG. 11C is 38, and the number of CIMs for S-R to H-R shown in FIG. 11D is 20.

12A is a graph of SS coefficients (N-R windows) of the training data set, and FIG. 12B is a graph of SS coefficients (A-R windows) of the training data set.

The details of CIM-based classification are as follows when a specific window is described as an example.

FIG. 12A shows a spectral subtracted (SS) coefficient included in the 17th window in the 11th time domain and frequency level.

FIG. 12A shows the values of 350 coefficients in the window of the N-R subtracted spectrogram obtained from the normal signal, and FIG. 12B shows the same values of the same window of the A-R subtracted spectrogram from the abnormal signal.

The red line is a threshold of 0.0089 as a result of learning.

An apparatus and method for diagnosing system health based on CIM according to the present invention described above can use CIM (Critical Information Map) to find an area including signals affecting failures in Time Frequency Representations (TFRs). It is done.

The present invention can provide a signal processing method applicable to a non-stationary signal using less data even without knowing internal information by using a CIM (Critical Information Map).

It will be understood that the present invention is implemented in a modified form without departing from the essential characteristics of the present invention as described above.

Therefore, the specified embodiments should be considered in terms of explanation rather than limitation, and the scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the equivalent range are included in the present invention. Should be interpreted.

100. Data synchronization unit
200. Subtract Spectrogram
300. Map value optimization unit

Claims (10)

A data synchronization unit using a cross-correlation as a method of extracting only signals of a corresponding frequency representing a non-stationary movement with a band pass filter and determining the degree of agreement between two types of time-series signals;
Subtraction Spectrogram using Spectral Subtraction that uses WPD (Wavelet Packet Decomposition) as a TFRs (Time Frequency Representations) method for CIM (Critical Information Map) representation, and inversely transforms the signal after spectralizing the signal. part;
CIM-based system health diagnosis, characterized in that it comprises; a map value optimizer for determining parameters for determining the location of the important region and the size of each region in the subtracted spectrogram (Subtracted Spectrogram) to create a CIM (Critical Information Map) Device for. According to claim 1, In the data synchronization unit,
In order to match the sync of the two signals by finding the delay value at which the value of the correlation between the two signals becomes the maximum at a specific sampling delay and applying it to the signal,
Certain motions in data with non-stationary signals

Find with,
Here, f is the original function g is a filtering function and τ is a device for diagnosing system health based on CIM, characterized in that it means a time delay of two signals. The method of claim 2, to synchronize the data,

And
Here, τ _max is the time delay when the value of cross-correlation is the maximum, and the Argmax function is an expression that derives the value of the domain in which the expression in parentheses becomes the maximum, and τ _max is equal to the raw signal. Device for diagnosing system health based on CIM, characterized by applying a delay. The method of claim 1, wherein the kernel function of the wavelet packet to decompose the signal into multiple frequencies in the WPD,

Is defined as,
A device for diagnosing system health based on CIM, wherein j is an index scale, k is a transformation operation, and n is a modulation parameter or a vibration parameter. The method of claim 4, wherein the scaling function

Defined as
Matrix function

Defined as,
function

And

Create
Here, h (k) and g (k) are orthogonal mirror filters orthogonal to each other. 6. The wavelet packet coefficients of states j, n, and k indicating the density of the filter,

Is defined as,
Here, f (t) is a device for diagnosing system health based on CIM, characterized in that it is a time signal to be analyzed. According to claim 1, Spectral subtraction (Spectral Subtraction) to inverse transform after the signal is spectrally subtracted by the subtraction spectrogram unit,

Defined as
Here, P _s (w) is a noise spectrum mixed with noise, P _n (w) is a noise spectrum measured in a silent state, and Fourier inverse transform of P ' _s (w) to obtain a noise-removed voice signal Device for diagnosing system health based on a CIM. The method of claim 7, wherein SS is extended to WT, which is one of the TFRs reflecting time, and expressed as an equation:

ego,
here,

Is the wavelet spectrum of the enhanced speech,

Is the wavelet spectrum of the observed signal,

A is a wavelet spectrum of noise, α is a reduction factor, a, b is a filter position and scale parameter for extracting one coefficient of a wavelet transform. . The method of claim 1, wherein the map value optimizer performs optimization by setting a target function to maximize a region selected for minimum information loss,
The first variable that determines the size of the region,
The second parameter is the threshold that is the basis for the density coefficient of one subtracted spectrogram to be considered as an effective value.
Device for diagnosing system health based on CIM, characterized in that a threshold for a number of valid values in a divided rectangle is selected as a third variable. Data Synchronization using Cross Correlation as a method of extracting only the signals of the corresponding frequency representing the non-stationary movement with a band pass filter and determining the degree of agreement between two types of time series signals ) step;
Subtraction Spectrogram using Spectral Subtraction that uses WPD (Wavelet Packet Decomposition) as a TFRs (Time Frequency Representations) method for CIM (Critical Information Map) representation, and inversely transforms the signal after spectralizing the signal. (Subtracted Spectrogram) generation step;
CIM (Critical Information Map) in order to create a Critical Information Map (Subtracted Spectrogram) in the subtracted spectrogram (Map Value Optimization) step of determining the parameters to determine the location of each region and the size of each area; CIM characterized in that it comprises a Method for diagnosis of system-based health.

Images