CN110653661A - Tool Condition Monitoring and Identification Method Based on Signal Fusion and Multifractal Spectrum Algorithm - Google Patents

Abstract

The invention relates to a tool state monitoring and identification method based on signal fusion and multifractal spectrum algorithm, comprising the following steps: step 1: collecting cutting force signals and vibration signals during the cutting process; step 2: collecting the cutting force signals and vibration signals in step 1 Signal, perform noise reduction processing, step 3: analyze the multifractal characteristics of the signal after the noise reduction, through the multifractal characteristics, find the connection between the signal and the tool wear, and extract the relevant multifractal spectrum from the calculated multifractal spectrum. Feature vector, use these feature vectors to represent the relationship between the signal and tool wear, step 4: Combine the feature vectors extracted in step 3 into a feature matrix as input parameter variables, build a support vector machine model for tool wear state monitoring, and use optimization The latter support vector machine model diagnoses the tool state for the unknown state, and the method of the present invention has a higher recognition rate for the tool state.

Description

Tool Condition Monitoring and Identification Method Based on Signal Fusion and Multifractal Spectrum Algorithm

technical field

The invention relates to the technical field of machining and advanced manufacturing, in particular to a tool state monitoring and identification method based on signal fusion and multifractal spectrum algorithm.

Background technique

Tool status intelligent monitoring technology is a fusion of technologies including sensor application, identification and analysis of digital signals, computer programming, and artificial intelligence machine learning. It has a significant role in promoting machining automation and unmanned manufacturing processes. field occupies an increasingly important position.

The current intelligent monitoring of tool status mostly uses a single signal for monitoring, including cutting force signal, vibration signal, acoustic emission signal and other signals. Although the single signal monitoring has the advantages of simple and easy operation, it has more disadvantages. Each signal can reflect the current tool wear state in the machining process in a certain way, but it is also too one-sided to reflect the current tool state comprehensively, which is easy to cause misjudgment of the tool wear state and easy to be processed. parameters, machine stiffness, workpiece material properties, and the influence of ambient noise. Use a variety of sensors to collect different signals, map the current tool status from different angles, make full use of the advantages and characteristics of different signals, and comprehensively reflect the tool information. Therefore, it is essential to propose a signal fusion technology for tool condition monitoring.

The current tool state can only be monitored. Many people extract the time domain features and frequency domain features of the signal, including mean, variance, skewness, kurtosis, center of gravity frequency, frequency variance, etc. Time domain analysis and frequency domain analysis are the most commonly used analysis methods for signal analysis. The information of the signal is analyzed from two different angles. The inventor found that most of the extracted features are based on the feature quantities obtained by statistics, although they are useful in the field of fault diagnosis. However, for the state monitoring in the field of tool wear, it is impossible to accurately establish the relationship between the signal and the tool wear by only using these features for various related signals generated in the cutting process, and then judge the tool state. Therefore, it is urgent to analyze the internal mechanism of the cutting process, to study the uniqueness of the signal generated by the cutting process, and to extract the signal characteristics through this uniqueness to establish the relationship between the signal and the tool wear.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a tool state monitoring and identification method based on signal fusion and multifractal spectrum algorithm to overcome the deficiencies of the prior art, with good identification effect and high identification rate.

To achieve the above object, the present invention adopts the following technical solutions:

The tool condition monitoring and identification method based on signal fusion and multifractal spectrum algorithm includes the following steps:

Step 1: Collect the cutting force signal and vibration signal of the tool during the cutting process.

Step 2: Perform noise reduction processing on the cutting force signal and vibration signal collected in Step 1 by using the wavelet threshold noise reduction method.

Step 3: Use the multifractal detrend fluctuation analysis method to analyze the signal sequence after denoising, analyze the multifractal characteristics of the signal, find the connection between the signal and the tool wear through the multifractal characteristics, and obtain the multifractal spectrum from the calculation. Extract the relevant eigenvectors and use these eigenvectors to characterize the relationship between the signal and tool wear.

Step 4: Combine the eigenvectors extracted in step 3 into a eigenmatrix as an input parameter variable, use the parameter optimization method to determine the important parameters in the model, build a support vector machine model for tool wear state monitoring, and optimize the support vector machine model , using the optimized support vector machine model to diagnose the tool state for the unknown state.

Further, the specific steps of the step 2 are:

Step (1): For the collected cutting force signal and vibration signal, use the set wavelet basis function and the set number of decomposition layers to do wavelet decomposition, and obtain signals of different frequency segments, including low-frequency signals and different numbers of high frequency signal.

Step (2): use the set threshold and the set threshold function to perform threshold quantization processing on the high-frequency segmented cutting force and vibration signals obtained by wavelet decomposition, remove signals higher than the threshold, and retain signals lower than the threshold.

Step (3): Perform wavelet reconstruction on the signal components obtained after threshold quantization together with the signal components in the low frequency band to obtain the vibration signal of the cutting force signal after wavelet noise reduction.

Further, in the step (1), the db3 wavelet basis function is selected to perform 4-layer wavelet decomposition on the signal.

Further, the specific steps of the step 3 are:

Step (a): For a time series { _xi , i=1, 2, ... N} of length N, calculate the difference y(i) between the value of x _i and the mean value, where x _i is the collected cutting force value or vibration amplitude.

Step (b): Divide the sequence y(i) into m subsequences according to the direction from the head end to the tail end, divide y(i) into m subsequences according to the direction from the tail end to the head end, and the subsequence length is s when dividing, Get 2m subsequences, m=int(N/s)

Step (c): Use the method of least squares to fit the local trend of each subsequence:

y _v (i)=a ₀ +a ₁ i+a ₂ i ² +...+ _ak i ^k , i=1,2,...; k=1,2,...

Among them, k is the polynomial order, and a _k is the polynomial coefficient;

Step (d): Calculate the mean square error function according to the results of steps (a) and (c):

When v=1,2,...m,

When v=m+1, m+2...2m,

Step (e): According to the result of step (d), calculate the q-order wave function:

Step (f): According to step (e), the power-law relationship between the q-order wave function F _q (s) and the subsequence length s is obtained: F _q (s)～sh ^(q) , and the generalized Hurst exponent is obtained h(q).

Step (g): According to the generalized Hurst index obtained in step (f), the relationship between the singularity index and the multifractal spectrum is obtained.

Further, the concrete method of described step (a) is:

in,

Further, the specific steps of the step (f) are: obtaining a certain linear relationship between the q-order wave function and the subsequence length s in the double logarithmic function coordinate system, and the slope is the generalized Hurst exponent h(q) .

Further, the specific steps of the step (g) are: there is the following relationship between the generalized Hurst exponent h(q) and the quality index τ(q):

τ(q)=qh(q)-1

The relationship between the singularity index α and the multifractal spectrum f(α) can be obtained through Legendre transformation:

α=τ′(q)=h(q)+qh′(q)

f(α)=qα-τ(q)

Further, in the step (3), the extracted feature vector is: the multifractal spectrum feature vector corresponding to each component vector of the cutting force signal and the vibration signal: (α _min , f(α _min ), α _max , f (α _max ), α ₀ , Δα, Δf(α)).

Among them, α _min represents the minimum value of α, α _max represents the maximum value of α, α ₀ is the corresponding value of α when f(α) is maximum, Δα=α _max -α _min , Δf(α)=f(α _max )-f(α _min ).

Beneficial effects of the present invention:

1. The present invention adopts the multi-fractal detrending fluctuation analysis method to analyze the signal, which has local dynamic characteristics and multi-analysis characteristics, and extracts the multi-fractal spectrum parameters as the feature vector, which can more comprehensively characterize the characteristics of the cutting signal, mainly in the local dynamics. Characteristics and multiple analysis characteristics, with a comprehensive feature description of the signal, the obvious difference between the initial state of the tool, the normal wear state and the sharp wear state is achieved, and a high recognition effect is obtained in the support vector machine model, and the recognition rate reaches more than 90%.

2. The present invention adopts the wavelet noise reduction method for the collected cutting force signal and vibration signal to remove noise, so as to obtain better signal characteristics. The db3 wavelet base function is selected to perform 4-layer wavelet decomposition on the signal, and the heuristic threshold is selected to perform threshold quantization on the high-frequency coefficients, and then wavelet reconstruction is performed to remove the signal noise and obtain better signal characteristics.

Description of drawings

The accompanying drawings that constitute a part of the present application are used to provide further understanding of the present application, and the schematic embodiments and descriptions of the present application are used to explain the present application and do not constitute a limitation to the present application.

1 is a schematic flowchart of Embodiment 1 of the present invention;

Fig. 2 is the cutting force signal waveform diagram after noise reduction in the initial wear state of the tool of the present invention;

3 is a waveform diagram of the cutting force signal after noise reduction under the normal wear state of the tool of the present invention;

Fig. 4 is the cutting force signal waveform diagram after noise reduction under the sharp wear state of the tool of the present invention;

Fig. 5 is the waveform diagram of the vibration signal after noise reduction in the initial wear state of the tool of the present invention;

Fig. 6 is the vibration signal waveform diagram after noise reduction under the normal wear state of the tool of the present invention;

7 is a waveform diagram of the vibration signal after noise reduction under the sharp wear state of the tool of the present invention;

8 is a graph showing the relationship between the q-order waveform function of the cutting force signal and the double logarithmic function of the subsequence length under the initial wear state of the tool of the present invention;

9 is a graph showing the relationship between the q-order waveform function of the cutting force signal and the double logarithmic function of the subsequence length under the normal wear state of the tool of the present invention;

10 is a graph showing the relationship between the q-order waveform function of the cutting force signal and the double logarithmic function of the subsequence length under the sharp wear state of the tool of the present invention;

11 is a graph showing the relationship between the q-order waveform function of the vibration signal and the double logarithmic function of the subsequence length under the initial wear state of the tool of the present invention;

12 is a graph showing the relationship between the q-order waveform function of the vibration signal and the double logarithmic function of the subsequence length under the normal wear state of the tool of the present invention;

13 is a graph showing the relationship between the q-order waveform function of the vibration signal and the double logarithmic function of the subsequence length under the sharp wear state of the tool of the present invention;

Fig. 14 is the relation diagram of generalized Hurst exponent and q of the cutting force signal of the present invention under different tool states;

Fig. 15 is the relation diagram of generalized Hurst index and q of vibration signal of the present invention under different tool states;

16 is a graph showing the relationship between the singularity index and the multifractal spectrum of the cutting force signal in the x-direction of the tool of the present invention in different states;

17 is a graph showing the relationship between the singularity index and the multifractal spectrum of the cutting force signal in the y direction of the tool of the present invention in different states;

18 is a graph showing the relationship between the singularity index and the multifractal spectrum of the cutting force signal in the z direction of the tool of the present invention under different states;

19 is a graph showing the relationship between the singularity index and the multifractal spectrum of the vibration signal of the tool of the present invention in the x-direction under different states;

20 is a graph showing the relationship between the singularity index and the multifractal spectrum of the vibration signal of the tool of the present invention in the y direction under different states;

21 is a graph showing the relationship between the singularity index and the multifractal spectrum of the vibration signal of the tool of the present invention in the z direction under different states;

Fig. 22 is the result diagram of SVM model tool recognition state after the completion of step 4 of the present invention;

Detailed ways

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

It should be noted that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural as well, furthermore, it is to be understood that when the terms "comprising" and/or "including" are used in this specification, it indicates that There are features, steps, operations, devices, components and/or combinations thereof.

For the convenience of description, if the words "up", "down", "left" and "right" appear in the present invention, it only means that the directions of up, down, left and right are consistent with the drawings themselves, and do not limit the structure. It is for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore should not be construed as a limitation of the present invention.

As described in the background art, the feature vectors collected by the existing tool state monitoring and identification methods cannot represent the comprehensive characteristics of the signal, and the identification effect is poor. In view of the above problems, the present application proposes a tool state monitoring method based on signal fusion and multifractal spectrum algorithm. recognition methods.

In Embodiment 1 of a typical implementation of the present application, as shown in FIG. 1 , the tool state monitoring and identification method based on signal fusion and multifractal spectrum algorithm includes the following steps:

Step 1: During the cutting process, the cutting force signal is collected by the dynamometer installed under the tool holder, and the vibration signal is collected by the acceleration sensor installed on the tool.

Step 2: For the cutting force signal and vibration signal collected in step 1, use the wavelet threshold noise reduction method to perform noise reduction processing to achieve the purpose of removing cutting noise, making the effective signal of the cutting process more obvious, and reducing other factors. The effect of subsequent signal analysis.

The specific steps of the step 2 are:

Step (1): 4-layer wavelet decomposition is performed on the cutting force signal and the vibration signal by using the db3 wavelet basis function.

Step (2): Use an appropriate threshold and an appropriate threshold function to perform threshold quantization on the high-frequency segmented cutting force and vibration signals obtained by wavelet decomposition, that is, remove signals higher than the threshold, and retain signals below the threshold.

Step (3): Perform wavelet reconstruction on the signal components obtained after threshold quantization together with the signal components in the low frequency band to obtain the vibration signal of the cutting force signal after wavelet noise reduction.

The above-mentioned steps (1) to (3) adopt the existing signal noise reduction processing method, and the detailed process thereof will not be described in detail here.

Figure 2 is the cutting force signal collected after noise reduction in the initial wear state of the tool, Figure 3 is the cutting force signal collected after noise reduction in the normal wear state of the tool, and Figure 4 is the cutting force signal collected after noise reduction in the sharp wear state of the tool .

Figure 5 is the vibration signal collected after noise reduction in the initial wear state of the tool, Figure 6 is the vibration signal collected after noise reduction in the normal wear state of the tool, and Figure 7 is the vibration signal collected after noise reduction in the sharp wear state of the tool.

Step 3: Perform multi-fractal analysis on the noise-reduced cutting force signal and vibration signal. After the multi-fractal analysis, extract the multi-fractal spectrum parameters to form feature vectors.

The step 3 includes the following specific steps:

Step (a): For a time series of length N { _xi , i=1, 2,...N}, where N is a natural number, calculate the cumulative dispersion value y(i) of x _i , where x _i is the collected cutting force value or vibration amplitude (including the values in the x, y, and z directions).

in,

Among them, n=1, 2...N

Step (b): Divide the sequence y(i) into m non-overlapping subsequences of length s according to the direction from the head end to the tail end, m=int(N/s), where s is a natural number, after division, There is a subsequence whose length is not s at the end of the sequence, so in order not to lose information, the sequence y(i) is subdivided into m (m=int(N/s)) non-overlapping lengths in the direction from the tail to the head. is a subsequence of s, and a total of 2m subsequences are obtained, where m is a natural number.

Step (c): Use the least squares method to fit the local trend of each subsequence:

y _v (i)=a ₀ +a ₁ i+a ₂ i ² +...+ _ak i ^k , i=1,2,...; k=1,2,...(3)

Among them, k is the polynomial order, and a _k is the polynomial coefficient;

Step (d): According to the results of step (a) and step (c), calculate the mean square error function: when v=1,2,...m,

When v=m+1, m+2...2m,

According to the result of step (d), calculate the q-order wave function:

Step (f): According to step (e), the power-law relationship between the q-order wave function F _q (s) and the subsequence length s is obtained: F _q (s)～sh ^(q) , and the generalized Hurst exponent is obtained h(q).

The specific method is: to obtain the linear relationship between the q-order wave function and the subsequence length s in the double logarithmic function coordinate system, then its slope is the generalized Hurst exponent h(q).

Figure 8 is a linear relationship diagram of the cutting force signal subsequence length s and the q-order wave function in the double logarithmic function coordinate system when q=10, 0, -10 when the tool is in the initial wear state.

Figure 9 is a linear relationship diagram of the cutting force signal subsequence length s and the q-order wave function in a double logarithmic function coordinate system when q=10, 0, -10 when the tool is in a normal wear state.

Figure 10 is a linear relationship diagram of the cutting force signal subsequence length s and the q-order wave function in the double logarithmic function coordinate system when the tool is in a state of rapid wear when q=10, 0, and -10.

Figure 11 is a linear relationship diagram of the vibration signal subsequence length s and the q-order wave function in the double logarithmic function coordinate system under the initial wear state of the tool when q=10, 0, and -10.

Figure 12 is a linear relationship diagram of the vibration signal subsequence length s and the q-order wave function in the double logarithmic function coordinate system when the tool is in a normal wear state when q=10, 0, and -10.

Figure 13 is a linear relationship diagram of the vibration signal subsequence length s and the q-order wave function in a double logarithmic function coordinate system when the tool is in a state of rapid wear when q=10, 0, and -10.

It can be seen from Figure 8-13 that the subsequence length s and the double logarithmic function of the q-order wave function have an approximate linear relationship. Under the set q value, a slope value can be obtained, that is, the generalized Hurst exponent h (q).

Taking different q values can get different h(q) values, and then get the relationship between the generalized Hurst exponent and q

As shown in FIG. 14 , the relationship between the generalized Hurst index of the cutting force signal and q is shown in FIG. 15 , and the relationship between the generalized Hurst index of the vibration signal and q is shown in FIG. 15 .

In step (g), the relationship between the singularity index and the multifractal spectrum is obtained according to the generalized Hurst index obtained in step (f).

The specific steps of the step (g) are: there is the following relationship between the generalized Hurst exponent h(q) and the quality index τ(q):

τ(q)=qh(q)-1 (8)

The relationship between the singularity index α and the multifractal spectrum f(α) can be obtained through Legendre transformation:

α=τ'(q)=h(q)+qh'(q) (9)

f(α)=qα-τ(q)

Then, the relationship between the singularity index α of the cutting force signal in the x, y and z directions and the multifractal spectrum f(α) is obtained, and the singularity index α of the vibration signal in the x, y and z directions and the multifractal spectrum f( α), as shown in Figure 16-21.

Extract the multifractal spectral eigenvectors corresponding to the components of the cutting force signal in the three directions and the components of the vibration signal in the three directions: (α _min , f(α _min ), α _max , f(α _max ), α ₀ , △α , △f(α)) has a total of 7x6 42-dimensional feature vectors.

Among them, α _min represents the minimum value of α, α _max represents the maximum value of α, α ₀ is the corresponding value of α when f(α) is maximum, Δα=α _max -α _min , Δf(α)=f(α _max )-f(α _min ).

Step 4: The 42-dimensional feature vector extracted in step (g) is input into the support vector machine model as a data set, and the tool state identification and diagnosis are carried out. The method of using the support vector machine model to identify and diagnose the tool state can adopt the existing method, In this embodiment, for a total of 270 samples of 42-dimensional feature vectors, 180 samples are randomly selected as a training set, and the remaining 90 samples are used as a test set.

The training set is input into the support vector machine model to train the model parameters. The training set and the test set are normalized first. Specifically, the data is scaled according to the proportion, so that the data with different sizes in different intervals can be in the original size order and The regularities fall within the same numerical interval, usually the interval [0,1]. In this embodiment, the data is normalized, that is, the data is uniformly mapped to the [0,1] interval. This method normalizes the data according to the mean and standard deviation of the original data. The processed data conform to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1. The transformation function is: x*=x-μσ where μ is the mean of all sample data, σ is the standard deviation of all sample data, and then Use the grid search method to train the two parameters c and g of the model, and finally get the parameters of the optimal model c=4, g=32; c is the penalty coefficient, which is the preference of the two important indicators of adjusting the interval size and classification accuracy The weight of , that is, the tolerance for errors. The higher the c, the less tolerance for errors, and the over-fitting situation is prone to occur, that is, there are too many own characteristics of the learning data set. The smaller the c, the easier the under-fitting situation occurs Without fully learning the features in the data sample, the value of c determines the size of the generalization ability. g:gamma is a parameter that comes with the RBF function after it is selected as the kernel. It determines the distribution of data after mapping from low-dimensional space to new high-dimensional feature space. The number of support vectors affects the speed of training and prediction. The obtained optimized model is input into the test set for testing. The above testing method adopts the analysis method of the existing support machine vector machine model, and the detailed process thereof will not be described in further detail.

As shown in Figure 22, the ordinate 1 represents the initial wear state, 2 represents the normal wear state, and 3 represents the sharp wear state. Finally, the tool state recognition accuracy is 92.2%.

In this embodiment, the initial wear state means that when the tool is just put into use, the actual contact area between the flank and the workpiece is very small, and the unit area bears a relatively large positive pressure. In addition, the flank of the newly sharpened tool is rough Large degree, microscopic unevenness, and rapid tool wear, this stage is called the initial wear stage of the tool

The normal wear state means that after the initial wear stage, the contact area between the flank and the workpiece increases, the pressure on the unit area gradually decreases, and the rough surface of the tool has gradually become smooth, and then the wear will enter a relatively stable normal state. wear period. In this stage, with the increase of cutting time, the wear amount of the flank face can be regarded as an approximately proportional increase, the wear is slow and uniform, and the duration is long, which is the effective working range of the tool

The sharp wear state means that after the tool wears evenly and gradually, the width of the wear band reaches a certain limit value. After the wear exceeds this value, the interaction between the tool and the workpiece is strengthened, the cutting force and cutting temperature rise sharply, the surface roughness of the part becomes larger, and the tool wears out. The increased speed is called the sharp wear phase.

Although the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, they do not limit the scope of protection of the present invention. Those skilled in the art should understand that on the basis of the technical solutions of the present invention, those skilled in the art do not need to pay creative efforts. Various modifications or deformations that can be made are still within the protection scope of the present invention.

Claims (8)

1. the tool state monitoring and identification method based on signal fusion and multifractal spectrum algorithm, is characterized in that, comprises the following steps: Step 1: Collect the cutting force signal and vibration signal of the tool during the cutting process; Step 2: Perform noise reduction processing on the cutting force signal and vibration signal collected in Step 1 by using the wavelet threshold noise reduction method; Step 3: Use the multifractal detrend fluctuation analysis method to analyze the signal sequence after denoising, analyze the multifractal characteristics of the signal, find the connection between the signal and the tool wear through the multifractal characteristics, and obtain the multifractal spectrum from the calculation. Extract relevant eigenvectors, and use these eigenvectors to characterize the relationship between the signal and tool wear; Step 4: Combine the eigenvectors extracted in step 3 into a eigenmatrix as an input parameter variable, use the parameter optimization method to determine the important parameters in the model, build a support vector machine model for tool wear state monitoring, and optimize the support vector machine model , using the optimized support vector machine model to diagnose the tool state for the unknown state. 2. the tool state monitoring and identification method based on signal fusion and multifractal spectrum algorithm as claimed in claim 1, is characterized in that, the concrete steps of described step 2 are: Step (1): For the collected cutting force signal and vibration signal, use the set wavelet basis function and the set number of decomposition layers to do wavelet decomposition, and obtain signals of different frequency segments, including low-frequency signals and different numbers of high frequency signal; Step (2): adopt the threshold value of setting and the threshold value function of setting to do threshold quantization processing to the high-frequency segmented cutting force and vibration signal obtained by wavelet decomposition, remove the signal higher than the threshold value, and retain the signal lower than the threshold value; Step (3): Perform wavelet reconstruction on the signal components obtained after threshold quantization together with the signal components in the low frequency band to obtain the vibration signal of the cutting force signal after wavelet noise reduction. 3 . The tool state monitoring and identification method based on signal fusion and multifractal spectrum algorithm as claimed in claim 2 , wherein, in the step (1), db3 wavelet basis function is selected to perform 4-layer wavelet decomposition on the signal. 4 . 4. the tool condition monitoring identification method based on signal fusion and multifractal spectrum algorithm as claimed in claim 1, is characterized in that, comprises the following steps: Step (a): For a time series { _xi , i=1, 2, ... N} of length N, calculate the difference y(i) between the value of x _i and the mean value, where x _i is the collected cutting force value or vibration amplitude; Step (b): Divide the sequence y(i) into m subsequences according to the direction from the head end to the tail end, divide y(i) into m subsequences according to the direction from the tail end to the head end, and the subsequence length is s when dividing, Obtain 2m subsequences, m=int(N/s); Step (c): Use the least squares method to fit the local trend of each subsequence: y _v (i)=a ₀ +a ₁ i+a ₂ i ² +...+ _ak i ^k , i=1,2,...; k=1,2,... Among them, k is the polynomial order, and a _k is the polynomial coefficient; Step (d): Calculate the mean square error function according to the results of steps (a) and (c): When v=1,2,...m,

When v=m+1, m+2...2m,

Step (e): According to the result of step (d), calculate the q-order wave function:

Step (f): According to step (e), the power-law relationship between the q-order wave function F _q (s) and the subsequence length s is obtained: F _q (s)～sh ^(q) , and the generalized Hurst exponent is obtained h(q); Step (g): According to the generalized Hurst index obtained in step (f), the relationship between the singularity index and the multifractal spectrum is obtained. 5. the tool state monitoring and identification method based on signal fusion and multifractal spectrum algorithm as claimed in claim 4, is characterized in that, the concrete method of described step (a) is:

in,

6. the tool condition monitoring and identification method based on signal fusion and multifractal spectrum algorithm as claimed in claim 4, is characterized in that, the concrete step of described step (f) is: obtain q-order wave function and subsequence length s in double The linear relationship in the logarithmic function coordinate system, the slope of which is the generalized Hurst exponent h(q). 7. The tool condition monitoring and identification method based on signal fusion and multifractal spectrum algorithm as claimed in claim 4, wherein the concrete steps of the step (g) are: generalized Hurst exponent h (q) and quality The following relationship exists between the exponents τ(q): τ(q)=qh(q)-1; The relationship between the singularity index α and the multifractal spectrum f(α) can be obtained by Legendre transformation: α=τ′(q)=h(q)+qh′(q) f(α)=qα-τ(q). 8. the tool condition monitoring and identification method based on signal fusion and multifractal spectrum algorithm as claimed in claim 4, is characterized in that, in described step 3, the characteristic vector of extraction is: each fraction of cutting force signal and vibration signal. The multifractal spectral feature vector corresponding to the vector (α _min , f(α _min ), α _max , f(α _max ), α ₀ , △α, △f(α)); Among them, α _min represents the minimum value of α, α _max represents the maximum value of α, α ₀ is the corresponding value of α when f(α) is maximum, Δα=α _max -α _min , Δf(α)=f(α _max )-f(α _min ).

Images