Acoustic vibration signal two-stage fusion-based gear milling machine disc type milling cutter wear state classification method

Abstract

The invention provides a gear milling machine disc type milling cutter wear state classification method based on sound and vibration signal two-stage fusion, and relates to the field of multi-source information fusion and fault diagnosis, and the method mainly comprises the steps: 1) employing four noise sensors to collect noise signals in four directions, obtaining the correlation degree of signals in different directions through a gray B-type correlation degree method, and obtaining the correlation degree of the signals in different directions; and performing data-level fusion to obtain a fused sound signal. 2) collecting a vibration signal of a main shaft of the gear milling machine, preprocessing the collected vibration signal and the fused sound signal, performing time-frequency feature extraction, performing optimization on extracted feature values by adopting a ReliefF-mRMR joint algorithm, and selecting features with relatively large influence to construct a data set; and 3) performing feature extraction on the data set by adopting BiLSTM, and classifying extracted feature vectors through XGBoost to obtain classification results of different wear states of the disc type milling cutter. According to the method, a two-stage fusion method is adopted, sound and vibration signals are fused, the characteristics of tool abrasion can be more effectively captured, and the tool abrasion state of the disc type milling cutter of the gear milling machine is judged.

Description

Technical Field

[0001] The invention relates to the fields of various parameter monitoring, data fusion and fault diagnosis of numerical control machine tools, and in particular to a wear state classification method for a disc-type milling cutter of a gear milling machine based on two-level fusion of acoustic vibration signals. Background Art

[0002] With the rapid development of modern manufacturing, CNC gear milling machines play an increasingly important role in the field of gear processing. As one of the core components of CNC gear milling machines, the wear state of the tool is directly related to the processing quality, efficiency and cost. The tool will inevitably wear out during the processing process, and the real-time monitoring and classification of the wear state is of great significance to ensure processing accuracy, extend tool life, reduce production costs and improve production safety. Therefore, it is very necessary to classify and monitor the wear state of the tool of the gear milling machine.

[0003] Multi-source signal fusion technology is an important technology in the field of information processing. It aims to extract effective and reliable information from information obtained from multiple different sources to improve the accuracy of decision-making and the performance of the system. With the rapid development of sensor technology, network communication technology, artificial intelligence and other fields, multi-source signal fusion technology has shown great application potential in many fields such as military, aerospace, transportation, medical care, and environmental monitoring. In modern scientific and technological applications, a single sensor or information source often cannot provide complete and comprehensive information. The fusion of multi-source information can make up for the shortcomings of a single information source and provide richer and more accurate data, thereby supporting more complex decision-making processes. Through the multi-level fusion of multi-source signals, the accuracy of tool wear state classification can be further improved. To this end, we propose a tool wear state classification method for disc-type milling cutters of gear milling machines based on two-level fusion of acoustic vibration signals. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a method for classifying the wear state of a disc-type milling cutter of a gear milling machine based on two-level fusion of acoustic vibration signals. The wear state of the disc-type milling cutter of a gear milling machine can be evaluated more accurately, effectively reducing the problem of low gear processing accuracy caused by tool wear.

[0005] To achieve the above purpose, the technical solution provided by the present invention is as follows:

[0006] A method for classifying tool wear states of a disc-type milling cutter of a gear milling machine based on two-level fusion of acoustic and vibration signals comprises the following steps:

[0007] A wear state classification method for a disc-type milling cutter of a gear milling machine based on two-level fusion of acoustic and vibration signals mainly includes the following steps:

[0008] S1: Four noise sensors are used to collect noise signals from four directions, and the correlation of signals from different directions is obtained by using the gray B-type correlation method, and the fused sound signal is obtained by data-level fusion;

[0009] S2: Use an acceleration sensor to collect the vibration signal of the milling machine spindle, then extract the time-frequency domain features of the fused sound signal and vibration signal, use the ReliefF-mRMR joint algorithm to optimize the feature vectors of the extracted features, and select the feature vectors with the greatest influence to construct a data set;

[0010] S3: BiLSTM is used to extract features from the data set, and the extracted features are transformed to form feature vectors that can be recognized by XGBoost. XGBoost is used for classification to obtain classification results of different wear states of gear milling machine tools.

[0011] In S1, four noise sensors are used to collect noise signals in four directions, and the correlation of signals in different directions is obtained by using the gray B-type correlation method, and data-level fusion is performed to obtain the fused sound signal, which mainly includes the following steps:

[0012] S1.1: The noise sensors are evenly arranged around the main spindle of the gear milling machine headstock, and the collected data are arranged in time series;

[0013] S1.2: Take one point as the reference sequence and the other three points as the comparison sequence to obtain the correlation between the reference sequence point and the other points; use the gray B-type correlation method to define the discrete signal as y i (k),y j (k), k=1,2,3...,n, first calculate the displacement difference:

[0014]

[0015] Then calculate the speed difference:

[0016] Then calculate the acceleration difference:

[0017] The formula of gray type B correlation is:

[0018]

[0019] Where X i is the reference sequence, X j To compare the sequences;

[0020] S1.3 After obtaining the gray B-type correlation of each point compared with other points, the correlation energy of the measuring point can be calculated, and then the relationship between the correlation and the weight can be used to obtain the sound fusion signal;

[0021] Assume that the gray B-type correlation energy of each point sound signal and the signals of the other three points is E ij , then the correlation signal energy of the i-th point is E i :

[0022]

[0023] S1.4 Since the correlation and weight are proportional, the sum of the weight values ​​is 1. The specific relationship is as follows:

[0024] E1:E2:E3:E4=P1:P2:P3:P4

[0025] P1+P2+P3+P4=1;

[0026] P1, P2, P3, and P4 are the weights of the four sound test points. After obtaining the weights of each point, weighted fusion is performed to obtain the sound fusion signal Nf fused at the data level:

[0027] Nf=P1X1+P2X2+P3X3+P4X4.

[0028] In the S2, an acceleration sensor is used to collect the vibration signal of the main shaft of the gear milling machine, and then the time-frequency domain features of the fused sound signal and the vibration signal are extracted. The feature vectors of the extracted features are optimized by the ReliefF-mRMR joint algorithm, and the feature vectors with great influence are selected to construct a data set, which mainly includes the following steps:

[0029] S2.1 uses an acceleration sensor placed at the spindle position. The tool is cut throughout its life and the vibration signal of each cut is collected. The vibration signals of the three wear states are marked as 1, 2, and 3 through the tool wear curve.

[0030] S2.2 extracts the time-frequency domain features of the fused sound signal and vibration signal, including time-domain features: mean, root mean square, standard deviation, waveform factor, skewness, kurtosis, peak, peak factor and impulse factor; frequency-domain features: power spectrum mean, frequency center of gravity, mean square frequency and time-frequency domain features, i.e. wavelet packet energy;

[0031] S2.3 First, the time-frequency features are screened by ReliefF, and a sample R is randomly selected from the sample set S. i , search with R i The k nearest neighbors H of the same category sample set j And the k nearest neighbors M of different categories of sample sets j , update the weight W of each feature according to the following formula:

[0032]

[0033] Where m is the number of iterations; diff(A,X,Y) represents the distance between two objects on feature A; k is the number of nearest neighbor samples; P(c) represents the prior probability of target category c;

[0034] According to W(A), useless feature sets are eliminated and 15 feature subsets are selected from 26 feature sets;

[0035] S2.4 After the optimization, the 15 feature subsets are screened again by the mRMR algorithm. If x and y are two continuous random variables, P(x) and P(y) are the probabilities corresponding to x and y respectively, and the joint probability is P(x, y), then the formula for the mutual information (MI) between the two is as follows:

[0036]

[0037] The formula of the mRMR principle is as follows:

[0038]

[0039] Where C is the target category; S is the feature set; is the number of feature quantities; x i 、x j represents the i-th and j-th features;

[0040] The mRMR principle is denoted as maxΦ(D,R). By using the mutual information quotient (MIQ) standard Φ(D,R) = D / R, the forward sequence search method is used to find the approximate optimal fault feature. Assuming X is the original feature set, there are m-1 feature subsets S m-1 , in the remaining features {XS m-1 The mth feature selected in} must satisfy:

[0041]

[0042] According to the results obtained by the m RMR principle, each feature is ranked from high to low, and the optimal feature classification subset is selected to construct the data set.

[0043] In the S3, BiLSTM is used to extract features from the data set, and the extracted features are converted to form feature vectors that can be recognized by XGBoost. XGBoost is used to classify and obtain classification results of different wear states of gear milling machine tools, which mainly includes the following steps:

[0044] S4.1 For the constructed data set, a bidirectional long-short term recurrent neural network BiLSTM is used to extract features from the data set. The final output calculation formula of BiLSTM is as follows:

[0045] hi =f1(w1x i +w2h i-1 );

[0046] H i =f2(w3x i +w5H i+1 );

[0047] Y i =f3(w4h i +w6H i );

[0048] Among them, x i (i=1,2,…,t) corresponds to the data input at time, h i (i=1,2,…,t) represents the hidden state of the LSTM in the forward iteration, H i (i=1,2,…,t) represents the LSTM hidden state of the backward iteration, Y i (i=1,2,…,t) represents the corresponding output data, w i (i=1,2,…,6) represents the weight of each layer, where f1,f2,f3 correspond to the activation functions of different layers;

[0049] S4.2 reconstructs the feature vector after BiLSTM feature extraction, and then enters the extreme gradient boosting decision tree XGBoost to classify the tool wear status.

[0050] The present invention provides a method for classifying wear states of disc-type milling cutters of gear milling machines based on two-level fusion of acoustic vibration signals, which has the following beneficial effects:

[0051] Compared with the judgment of tool wear status by a single sensor, multi-sensor fusion has higher accuracy in tool wear status classification. And compared with the first-level fusion that only fuses the feature layer, the two-level fusion further improves the judgment accuracy. The classification using XGBoost has higher accuracy than the classification using the softmax classifier. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is the distribution diagram of sensor measurement points.

[0053] Figure 2 This is the schematic diagram of the acquisition system.

[0054] Figure 3 Diagnostic model flow chart.

[0055] Figure 4 This is the BiLSTM network structure diagram. DETAILED DESCRIPTION

[0056] Specific embodiments of the present invention will be described with reference to the accompanying drawings and examples.

[0057] As attached Figure 1 , attached Figure 2 And attached Figure 3 As shown, a method for classifying tool wear states of a disc-type milling cutter of a gear milling machine based on two-level fusion of acoustic and vibration signals comprises the following steps:

[0058] S1: Four noise sensors are used to collect noise signals from four directions, and the correlation of signals from different directions is obtained by using the gray B-type correlation method, and the fused sound signal is obtained by data-level fusion;

[0059] S2: Use an acceleration sensor to collect the vibration signal of the milling machine spindle, then extract the time-frequency domain features of the fused sound signal and vibration signal, use the ReliefF-mRMR joint algorithm to optimize the feature vectors of the extracted features, and select the feature vectors with the greatest influence to construct a data set;

[0060] S3: BiLSTM is used to extract features from the data set, and the extracted features are transformed to form feature vectors that can be recognized by XGBoost. XGBoost is used for classification to obtain classification results of different wear states of gear milling machine tools.

[0061] Among them, as a possible implementation method, four noise sensors are used to collect noise signals in four directions, and the correlation of signals in different directions is obtained by the gray B-type correlation method, and data-level fusion is performed to obtain the fused sound signal, which mainly includes the following steps:

[0062] S1.1: The noise sensors are evenly arranged around the main spindle of the gear milling machine headstock, and the collected data are arranged in time series;

[0063] S1.2: Take one point as the reference sequence and the other three points as the comparison sequence to obtain the correlation between the reference sequence point and the other points. Using the gray B-type correlation method, the discrete signal is defined as y i (k),y j (k), k=1,2,3...,n, first calculate the displacement difference:

[0064]

[0065] Then calculate the speed difference:

[0066]

[0067] Then calculate the acceleration difference:

[0068] The formula of gray type B correlation is:

[0069]

[0070] Where X i is the reference sequence, X j To compare the sequence.

[0071] S1.3 After obtaining the gray B-type correlation degree of each point compared with other points, the correlation energy of the measurement point can be calculated, and then the relationship between the correlation degree and the weight can be used to obtain the sound fusion signal. Assume that the gray B-type correlation energy of the sound signal of each point and the signals of the other three points is E ij , then the correlation signal energy of the i-th point is E i :

[0072]

[0073] S1.4 Since the correlation and weight are proportional, the sum of the weight values ​​is 1. The specific relationship is as follows:

[0074] E1:E2:E3:E4=P1:P2:P3:P4

[0075] P1+P2+P3+P4=1;

[0076] P1, P2, P3, and P4 are the weights of the four sound test points. After obtaining the weights of each point, weighted fusion is performed to obtain the sound fusion signal Nf fused at the data level:

[0077] Nf=P1X1+P2X2+P3X3+P4X4;

[0078] Among them, as a possible implementation method, an acceleration sensor is used to collect the vibration signal of the spindle of the gear milling machine, and then the time-frequency domain features of the fused sound signal and vibration signal are extracted. The feature vectors of the extracted features are optimized through the ReliefF-mRMR joint algorithm, and the feature vectors with great influence are selected to construct the data set, which mainly includes the following steps:

[0079] S2.1 uses an acceleration sensor placed at the spindle position. The tool goes through the entire cutting life and collects the vibration signal of each cut. The vibration signals of the three wear states are marked as 1, 2, and 3 through the tool wear curve.

[0080] S2.2 extracts the time-frequency domain features of the fused sound signal and vibration signal, including time domain features (mean, root mean square, standard deviation, waveform factor, skewness, kurtosis, peak, peak factor, pulse factor), frequency domain features (power spectrum mean, frequency center of gravity, mean square frequency) and time-frequency domain features (wavelet packet energy).

[0081] S2.3 First, the time-frequency features are screened by ReliefF, and a sample R is randomly selected from the sample set S. i , search with R i The k nearest neighbors H of the same category sample set j And the k nearest neighbors M of different categories of sample sets j , update the weight W of each feature according to the following formula:

[0082]

[0083] Where m is the number of iterations; diff(A,X,Y) represents the distance between two objects on feature A; k is the number of nearest neighbor samples; P(c) represents the prior probability of target category c.

[0084] Useless feature sets are eliminated according to W(A), and 15 feature subsets are selected from the 26 feature sets.

[0085] S2.4 After the optimization, the 15 feature subsets are screened again by the mRMR algorithm. If x and y are two continuous random variables, P(x) and P(y) are the probabilities corresponding to x and y respectively, and the joint probability is P(x, y), then the formula for the mutual information (MI) between the two is as follows:

[0086]

[0087] The formula of the mRMR principle is as follows:

[0088]

[0089] Where C is the target category; S is the feature set; is the number of feature quantities; x i 、x j Denotes the i-th and j-th features. The mRMR principle is denoted as maxΦ(D,R). By using the mutual information quotient (MIQ) standard Φ(D,R) = D / R, the forward sequence search method is used to find the approximate optimal fault feature. Assuming X is the original feature set, there are m-1 feature subsets S. m-1 , in the remaining features {XS m-1 The mth feature selected in} must satisfy:

[0090]

[0091] According to the results obtained by the mRMR principle, each feature is ranked from high to low, and the optimal feature classification subset is selected to construct the data set.

[0092] Among them, as a possible implementation method, as shown in the attached Figure 4As shown in the figure, BiLSTM is used to extract features from the data set, and the extracted features are converted to form feature vectors that can be recognized by XGBoost. XGBoost is used for classification to obtain the classification results of different wear states of gear milling machine tools. The main steps include:

[0093] S4.1 For the constructed data set, a bidirectional long-short term recurrent neural network BiLSTM is used to extract features from the data set. The final output calculation formula of BiLSTM is as follows:

[0094] h i =f1(w1x i +w2h i-1 );

[0095] H i =f2(w3x i +w5H i+1 );

[0096] Y i =f3(w4h i +w6H i );

[0097] Among them, x i (i=1,2,…,t) corresponds to the data input at time, h i (i=1,2,…,t) represents the hidden state of the LSTM in the forward iteration, H i (i=1,2,…,t) represents the LSTM hidden state of the backward iteration, Y i (i=1,2,…,t) represents the corresponding output data, w i (i=1,2,…,6) represents the weight of each layer, where f1,f2,f3 correspond to the activation functions of different layers.

[0098] S4.2 reconstructs the feature vector after BiLSTM feature extraction, and then enters the extreme gradient boosting decision tree XGBoost to classify the tool wear status.

[0099] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only preferred examples of the invention and are not intended to limit the present invention. Without departing from the novel spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for classifying the wear state of a disc-type milling cutter of a gear milling machine based on two-level fusion of acoustic and vibration signals, characterized in that: The main steps are as follows: S1: Four noise sensors are used to collect noise signals from four directions, and the correlation of signals from different directions is obtained by using the gray B-type correlation method, and the fused sound signal is obtained by data-level fusion; S2: Use an acceleration sensor to collect the vibration signal of the gear milling machine spindle, then extract the time-frequency domain features of the fused sound signal and vibration signal, use the ReliefF-mRMR joint algorithm to optimize the feature vectors of the extracted features, and select the feature vectors with the greatest influence to construct a data set; S3: BiLSTM is used to extract features from the data set, and the extracted features are transformed to form feature vectors that can be recognized by XGBoost. XGBoost is used for classification to obtain classification results of different wear states of gear milling machine tools.

2. A method for classifying wear states of disc-type milling cutters of gear milling machines based on two-level fusion of acoustic and vibration signals according to claim 1, characterized in that: In S1, four noise sensors are used to collect noise signals in four directions, and the correlation of signals in different directions is obtained by using the gray B-type correlation method, and data-level fusion is performed to obtain the fused sound signal, which mainly includes the following steps: S1.1: The noise sensors are evenly arranged around the main spindle of the gear milling machine headstock, and the collected data are arranged in time series; S1.2: Take one point as the reference sequence and the other three points as the comparison sequence to obtain the correlation between the reference sequence point and the other points; use the gray B-type correlation method to define the discrete signal as y i (k),y j (k), k=1,2,3...,n, first calculate the displacement difference: Then calculate the speed difference: Then calculate the acceleration difference: The formula of gray type B correlation is: Where X i is the reference sequence, X j To compare the sequences; S1.3 After obtaining the gray B-type correlation of each point compared with other points, the correlation energy of the measuring point can be calculated, and then the relationship between the correlation and the weight can be used to obtain the sound fusion signal; Assume that the gray B-type correlation energy of each point sound signal and the signals of the other three points is E ij , then the correlation signal energy of the i-th point is E i : S1.4 Since the correlation and weight are proportional, the sum of the weight values ​​is 1. The specific relationship is as follows: E1:E2:E3:E4=P1:P2:P3:P4 P1+P2+P3+P4=1; P1, P2, P3, and P4 are the weights of the four sound test points. After obtaining the weights of each point, weighted fusion is performed to obtain the sound fusion signal Nf fused at the data level: Nf=P1X1+P2X2+P3X3+P4X4.

3. The wear state classification method of a gear milling machine disc cutter based on two-level fusion of acoustic vibration signals according to claim 1, characterized in that: In the S2, an acceleration sensor is used to collect the vibration signal of the main shaft of the gear milling machine, and then the time-frequency domain features of the fused sound signal and the vibration signal are extracted. The feature vectors of the extracted features are optimized by the ReliefF-mRMR joint algorithm, and the feature vectors with great influence are selected to construct a data set, which mainly includes the following steps: S2.1 uses an acceleration sensor placed at the spindle position. The tool is cut throughout its life and the vibration signal of each cut is collected. The vibration signals of the three wear states are marked as 1, 2, and 3 through the tool wear curve. S2.2 extracts the time-frequency domain features of the fused sound signal and vibration signal, including the time-domain features: mean, root mean square, standard deviation, waveform factor, skewness, kurtosis, peak value, peak factor and impulse factor; Frequency domain features: power spectrum mean, frequency center of gravity, mean square frequency and time-frequency domain features, i.e. wavelet packet energy; S2.3 First, the time-frequency features are screened by ReliefF, and a sample R is randomly selected from the sample set S. i , search with R i The k nearest neighbors H of the same category sample set j And the k nearest neighbors M of different categories of sample sets j , update the weight W of each feature according to the following formula: Where m is the number of iterations; diff(A,X,Y) represents the distance between two objects on feature A; k is the number of nearest neighbor samples; P(c) represents the prior probability of target category c; According to W(A), useless feature sets are eliminated and 15 feature subsets are selected from 26 feature sets; S2.4 After the optimization, the 15 feature subsets are screened again by the mRMR algorithm. If x and y are two continuous random variables, P(x) and P(y) are the probabilities corresponding to x and y respectively, and the joint probability is P(x, y), then the formula for the mutual information (MI) between the two is as follows: The formula of the mRMR principle is as follows: Where C is the target category; S is the feature set; is the number of feature quantities; x i 、x j represents the i-th and j-th features; The mRMR principle is denoted as maxΦ(D,R). By using the mutual information quotient (MIQ) standard Φ(D,R) = D / R, the forward sequence search method is used to find the approximate optimal fault feature. Assuming X is the original feature set, there are m-1 feature subsets S m-1， In the remaining features {XS m-1 The mth feature selected in} must satisfy: According to the results obtained by the m RMR principle, each feature is ranked from high to low, and the optimal feature classification subset is selected to construct the data set.

4. The wear state classification method of a gear milling machine disc cutter based on two-level fusion of acoustic vibration signals according to claim 1, characterized in that: In the S3, BiLSTM is used to extract features from the data set, and the extracted features are converted to form feature vectors that can be recognized by XGBoost. XGBoost is used to classify and obtain classification results of different wear states of gear milling machine tools, which mainly includes the following steps: S4.1 For the constructed data set, a bidirectional long-short term recurrent neural network BiLSTM is used to extract features from the data set. The final output calculation formula of BiLSTM is as follows: h i =f1(w1x i +w2h i-1 ); <h2 style=";text-align:left;direction:ltr">H<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> =f2(w3x<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> +w5H<h2 style=";text-align:left;direction:ltr"> i+1 <h2 style=";text-align:left;direction:ltr"> ); AND i =f3(w4h i +w6H i ); Among them, x i (i=1,2,…,t) corresponds to the data input at time, h i (i=1,2,…,t) represents the hidden state of the LSTM in the forward iteration, H i (i=1,2,…,t) represents the LSTM hidden state of the backward iteration, Y i (i=1,2,…,t) represents the corresponding output data, w i (i=1,2,…,6) represents the weight of each layer, where f1,f2,f3 correspond to the activation functions of different layers; S4.2 reconstructs the feature vector after BiLSTM feature extraction, and then enters the extreme gradient boosting decision tree XGBoost to classify the tool wear status.

Images