Cutter wear state monitoring method based on power sensor

Abstract

The invention discloses a tool wear state monitoring method based on a power sensor, and the method comprises the steps: collecting sample data of voltage and current values of a main shaft in a machining process of a machine tool in real time, converting the sample data into power signals, and recording a tool wear state corresponding to each piece of sample data; preprocessing the power signals, calculating time domain features and frequency domain features of the preprocessed power signals, extracting frequency band energy features corresponding to tool wear through fast Fourier transform and wavelet packet decomposition to jointly form a feature sample set, and encoding the feature sample set with corresponding tool state types to form a sample set; optimizing the BP neural network based on a grey wolf particle swarm algorithm; high-correlation features are screened out through a Pearson's correlation coefficient method, feature parameters serve as input variables of a BP neural network, the tool wear state serves as output variables, and the tool wear state is judged according to the screened feature parameters. And the recognition efficiency and the recognition precision of the tool wear state are effectively improved.

Description

Technical Field

[0001] The invention belongs to the technical field of tool state detection and sensor data acquisition in the mechanical field, and in particular relates to a tool wear state monitoring method based on a power sensor. Background Art

[0002] As one of the main processing methods in the manufacturing industry, cutting processing has a significant impact on improving manufacturing level and efficiency. As an indispensable part of cutting processing, monitoring the status of cutting tools, that is, tool wear, has become an indispensable step in the production process. Therefore, it is increasingly important to carry out status monitoring of the cutting process of CNC machine tool automation production lines. In view of the complex and changeable actual processing environment, the tool status can often only be monitored indirectly by collecting sensor signals and other methods. By collecting sensor signals, processing them, and extracting them, the current tool wear status monitoring methods mainly have the following problems: First, the effective judgment of complex processing environments is not high. If manual monitoring is required, a lot of experience is needed to judge, which has high requirements for workers; second, a large number of data samples are required to ensure recognition accuracy, resulting in low recognition efficiency and weak anti-interference strength. Summary of the invention

[0003] In view of the above technical problems, the present invention provides a tool wear state monitoring method based on a power sensor.

[0004] The technical solution adopted by the present invention to solve the technical problem is:

[0005] A method for monitoring tool wear state based on a power sensor, the method comprising the following steps:

[0006] S100: arranging a power sensor at the inverter input end of the machine tool, collecting sample data of the spindle voltage and current values ​​during the machine tool processing in real time through the sensor, and converting them into power signals, while recording the tool wear status corresponding to each sample data;

[0007] S200: preprocessing the power signal, calculating the time domain features and frequency domain features of the preprocessed power signal, extracting the energy features of the frequency band corresponding to tool wear through fast Fourier transform and wavelet packet decomposition, and jointly forming a feature sample set, encoding the feature sample set and the corresponding tool state type to form a sample set;

[0008] S300: Taking the characteristic parameters as input and the tool wear status as output, the BP neural network is optimized by using the sample set combined with the Grey Wolf Particle Swarm Optimization Algorithm to establish a tool wear status identification model based on the BP neural network;

[0009] S400: Use the Pearson correlation coefficient method to filter out features with correlation coefficients greater than a preset threshold from the feature sample set, complete the dimensionality reduction of the feature sample set, and use the filtered feature parameters as input variables of the tool wear state identification model, and the tool wear state as the output variable to determine the mapping relationship between the input and output, and judge the tool wear state based on the filtered feature parameters.

[0010] Preferably, S100 includes:

[0011] S110: presetting the tool running trajectory and processing parameters, collecting the current and voltage signals of the spindle at fixed intervals, observing the tool wear state with a microscope, measuring the tool wear amount, and calibrating the tool wear state at this time;

[0012] S120: Repeat the experiment in this way to obtain sample data of current and voltage values ​​of multiple spindles, and convert them into power signals. At the same time, the tool wear status corresponding to each sample data is recorded.

[0013] Preferably, the time domain features in S200 include mean, root mean square, standard deviation and peak-to-peak value, and the frequency domain features include frequency mean, frequency centroid and frequency standard deviation. The time domain features and frequency domain features of the preprocessed power signal are calculated in S200, specifically:

[0014] The mean is:

[0015]

[0016] The root mean square is:

[0017]

[0018] The standard deviation is:

[0019]

[0020] Where X(i) is the measured value of the power signal at the i-th moment, N is the total number of points of the power signal, and x is the average value of the power signal;

[0021] The peak-to-peak value is:

[0022] PP=max(x(i))-min(x(i))

[0023] Wherein, max(x(i)) is the maximum value of the power signal data, and min(x(i)) is the minimum value of the power signal;

[0024] The frequency mean is:

[0025]

[0026] The frequency center of gravity is specifically:

[0027]

[0028] The frequency standard deviation is:

[0029]

[0030] Where k is the total number of spectral lines, f(k) is the frequency value of the kth spectral line, and |X(k)| is the absolute value of the power spectrum value corresponding to the kth spectral line.

[0031] Preferably, in S200, the energy features of the frequency bands corresponding to tool wear are extracted by fast Fourier transform and wavelet packet decomposition to form a feature sample set, specifically:

[0032] S210: Performing fast Fourier transform on the power signal after noise reduction to convert the time domain signal into a frequency domain signal;

[0033] S220: Perform wavelet packet decomposition on the power signal after noise reduction using wavelet basis functions, decompose the signal into frequency bands, calculate the proportion of each frequency band energy to the total energy, and select the frequency band whose energy proportion exceeds a preset threshold as the effective frequency band; by analyzing the change trend of the energy of each effective frequency band under different tool wear conditions, select the frequency band energy that is highly correlated with the tool wear condition as the sample feature.

[0034] Preferably, the specific steps of wavelet packet decomposition in S220 are as follows:

[0035] S221: performing multi-layer decomposition on the power signal after noise reduction using wavelet basis functions to construct a wavelet packet tree; each layer of decomposition includes high-frequency and low-frequency filtering on the power signal after noise reduction, and then down-sampling the filtered signal;

[0036] S222: The power signal x(t) after noise reduction is passed through a low-pass filter h(t) and a high-pass filter g(t) to obtain approximate coefficients and detail coefficients respectively, and the approximate coefficients and detail coefficients are further subjected to low-pass and high-pass filtering to obtain coefficients of the next layer, and this process is repeated until a predetermined number of decomposition layers is reached;

[0037] S223: Calculate the square sum of the approximate coefficient and the detail coefficient of each frequency band to obtain the energy value of each frequency band, calculate the proportion of the energy of each frequency band to the total energy, and select the frequency band whose energy proportion exceeds a preset threshold as the effective frequency band.

[0038] Preferably, the approximate coefficient in S222 is specifically:

[0039]

[0040] The detail coefficients are:

[0041]

[0042] The approximate coefficients and detail coefficients of each layer are further decomposed:

[0043]

[0044] Among them, a j [n] is the approximation coefficient of the jth layer, dj[n] is the detail coefficient of the jth layer; h[k] is the low-pass filter coefficient, and g[k] is the high-pass filter coefficient; is the coefficient of the nth node in the jth layer.

[0045] Preferably, the frequency band energy calculation formula in S223 is:

[0046]

[0047] The energy ratio calculation formula is:

[0048]

[0049] Among them, E(f k ) is the energy of the kth frequency band, f k1 and f k2 are the starting and ending frequencies of the kth frequency band, and K is the total number of frequency bands.

[0050] Preferably, S300 includes:

[0051] S310: Input the sample set and randomly initialize the weight and bias of the BP neural network, initialize the position and speed of the particle swarm and the wolf pack, and initialize the training parameters; wherein the training parameters include the population size, inertia coefficient, individual learning factor, social learning factor and total number of iterations;

[0052] S320: Calculate the fitness value of each individual according to the fitness function and sort them from high to low, select the first three individuals with the best solutions according to the fitness value, and record their position vectors;

[0053] S330: Update the position and velocity of each particle using the Grey Wolf optimization algorithm;

[0054] S340: Calculate the fitness value of each individual after the update, and compare it with the value of the previous iteration, select the individuals with the top three best positions, and the position vectors of these three optimal solution individuals represent the optimal weights and biases in the current iteration;

[0055] S350: Use the current optimal weights and biases to train the BP neural network, calculate the training error, and continuously adjust the weights and biases until the termination condition is met, thereby completing the optimization of the BP neural network.

[0056] Preferably, S330 is specifically:

[0057] The position update formula of each particle is:

[0058] X(t+1)=X(t)+V(t+1)

[0059] The speed update formula is:

[0060] V(t+1)=ωV(t)+c1r1[P b -X(t)]+w α c2r 2α [P α -X(t)]+

[0061] w β c2r 2β [P β -X(t)]+w δ c2r 2δ [P δ -X(t)]

[0062] Among them, c1r1[P b -X(t)] is the relationship between the current velocity increment and the particle's best historical position; P α , P β , P δ are the position vectors of the individuals in the optimal solution of SEG, c1 is the individual learning factor, ω is the inertia coefficient, c2 is the social learning factor, r1 and r2 are random numbers between [0, 1]; 2α 、r 2β 、r 2δ are the social learning factors corresponding to the three wolves, r 2α 、r 2β 、r 2δ is a random number between [0, 1]; w α 、w β and w δ The weights corresponding to the three wolves are set to 0.45, 0.3, and 0.25 respectively.

[0063] Preferably, in S500, the features with correlation coefficients greater than a preset threshold value are screened from the feature sample set using the Pearson correlation coefficient method, including:

[0064] S510: Pearson correlation coefficient method is used to calculate the Pearson correlation coefficient between each feature in the sample feature set and the tool wear state, specifically:

[0065]

[0066] Among them, the Pearson correlation r is used to measure the linear correlation between two variables, and its value range is [-k, 1], X i and Y i are the i-th sample values ​​of the characteristic variable X and tool wear state Y, and are the means of X and Y respectively;

[0067] S520: Screen out features whose absolute values ​​are greater than a preset threshold and use them as input variables of a subsequent tool wear state identification model.

[0068] The above-mentioned tool wear state monitoring method based on power sensor first performs multi-domain analysis on the power signal, which can more accurately capture and understand the essence of the signal; at the same time, in view of the problems that BP neural network is prone to fall into local optimality and is sensitive to initial weights, the gray wolf particle swarm algorithm is used to optimize the weights and biases of the BP neural network to improve the network performance and prediction accuracy; finally, a tool wear state identification model based on BP neural network is established to accurately identify the tool wear state, effectively improving the recognition efficiency and recognition accuracy of the tool wear state. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a flow chart of a tool wear state monitoring method based on a power sensor in one embodiment of the present invention;

[0070] Figure 2 A BP neural network structure diagram provided in one embodiment of the present invention;

[0071] Figure 3 A flowchart of optimizing a BP neural network using a grey wolf particle swarm algorithm provided in one embodiment of the present invention;

[0072] Figure 4 It is a Pearson correlation coefficient diagram of some characteristic parameters after power signal processing provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0073] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings.

[0074] In one embodiment, Figure 1 As shown, a tool wear state monitoring method based on a power sensor comprises the following steps:

[0075] S100: arranging a power sensor at the inverter input end of the machine tool, collecting sample data of the spindle voltage and current values ​​during the machine tool processing in real time through the sensor, and converting them into power signals, while recording the tool wear status corresponding to each sample data;

[0076] S200: Preprocess the power signal, calculate the time-domain features and frequency-domain features of the preprocessed power signal, extract the energy features of the corresponding frequency bands of tool wear through fast Fourier transform and wavelet packet decomposition, jointly form a feature sample set, encode the feature sample set and the corresponding tool state types to form a sample set; further, divide the sample set into a training set and a test set according to a preset ratio.

[0077] S300: Take the feature parameters as the input and the tool wear state as the output, use the sample set to optimize the BP neural network in combination with the grey wolf particle swarm algorithm, and establish a tool wear state identification model based on the BP neural network.

[0078] S400: Use the Pearson correlation coefficient method to screen out the features with a correlation coefficient greater than the preset threshold from the feature sample set, complete the dimensionality reduction of the feature sample set, use the selected feature parameters as the input variables of the tool wear state identification model, the tool wear state as the output variable, determine the mapping relationship between the input and output, and judge the tool wear state according to the selected feature parameters.

[0079] In one embodiment, S100 includes:

[0080] S110: Preset the tool running trajectory and machining parameters in advance, collect the current and voltage signals of the spindle at fixed time intervals, observe the tool wear state with a microscope, measure the tool wear amount, and calibrate the tool wear state at this time.

[0081] S120: Conduct repeated experiments in this way to obtain sample data of the current and voltage values of multiple spindles, convert them into power signals, and record the tool wear state corresponding to each sample data at the same time.

[0082] Specifically, the milling experiment conditions are shown in the following table:

[0083]

[0084] The experiment adopts the above cutting conditions to repeat the full-life cycle wear experiment. Under the set cutting parameters, measure the tool wear amount on the premise of machining a fixed distance until the tool wears out. According to the flank wear VB of the tool, the tool wear state is divided into initial wear (VB ≤ 0.12 mm), normal wear (0.12 mm < VB ≤ 0.3 mm), and severe wear (VB > 0.3 mm). And organize the tool state types and their corresponding power signals obtained from the experiment into samples.

[0085] In one embodiment, the preprocessing of the power signal in S200 includes:

[0086] The wavelet threshold denoising method is used to perform wavelet denoising on the power signal collected in the experiment to improve the signal-to-noise ratio.

[0087] In one embodiment, the time domain features in S200 include mean, root mean square, standard deviation and peak-to-peak value, and the frequency domain features include frequency mean, frequency center of gravity and frequency standard deviation. The time domain features and frequency domain features of the preprocessed power signal are calculated in S200, specifically:

[0088] The mean is:

[0089]

[0090] The root mean square is:

[0091]

[0092] The standard deviation is:

[0093]

[0094] Where X(i) is the measured value of the power signal at the i-th moment, N is the total number of points of the power signal, and x is the average value of the power signal;

[0095] The peak-to-peak value is:

[0096] PP=max(x(i))-min(x(i))

[0097] Wherein, max(x(i)) is the maximum value of the power signal data, and min(x(i)) is the minimum value of the power signal;

[0098] The frequency mean is:

[0099]

[0100] The frequency center of gravity is specifically:

[0101]

[0102] The frequency standard deviation is:

[0103]

[0104] Where k is the total number of spectral lines, f(k) is the frequency value of the kth spectral line, and |X(k)| is the absolute value of the power spectrum value corresponding to the kth spectral line.

[0105] Furthermore, the mean represents the average value of the signal, the root mean square represents the average energy of the signal, the standard deviation represents the fluctuation range of the signal, the peak-to-peak value represents the range of variation of the signal, the frequency mean represents the average value of the amplitudes of different frequency components of the signal in the frequency domain, the frequency center of gravity represents the frequency corresponding to the peak with a larger component in the spectrum, and the frequency standard deviation represents the degree of discreteness or concentration of the signal spectrum.

[0106] In one embodiment, in S200, the energy characteristics of the frequency band corresponding to tool wear are extracted by fast Fourier transform and wavelet packet decomposition to form a feature sample set, which is specifically:

[0107] S210: Performing fast Fourier transform on the power signal after noise reduction to convert the time domain signal into a frequency domain signal;

[0108] S220: Perform wavelet packet decomposition on the denoised power signal using wavelet basis functions, decompose the signal into frequency bands, calculate the proportion of each frequency band energy to the total energy, and select frequency bands whose energy proportion exceeds a preset threshold (3%) as valid frequency bands; by analyzing the changing trend of the energy of each valid frequency band under different tool wear states, select the frequency band energy that is highly correlated with the tool wear state as the sample feature.

[0109] The specific steps of wavelet packet decomposition in S220 are as follows:

[0110] S221: performing multi-layer decomposition on the power signal after noise reduction using wavelet basis functions to construct a wavelet packet tree; each layer of decomposition includes high-frequency and low-frequency filtering on the power signal after noise reduction, and then down-sampling the filtered signal;

[0111] S222: The power signal x(t) after noise reduction is passed through a low-pass filter h(t) and a high-pass filter g(t) to obtain approximate coefficients and detail coefficients respectively, and the approximate coefficients and detail coefficients are further subjected to low-pass and high-pass filtering to obtain coefficients of the next layer, and this process is repeated until a predetermined number of decomposition layers is reached;

[0112] S223: Calculate the square sum of the approximate coefficient and the detail coefficient of each frequency band to obtain the energy value of each frequency band, calculate the proportion of the energy of each frequency band to the total energy, and select the frequency band whose energy proportion exceeds a preset threshold as the effective frequency band.

[0113] In one embodiment, the approximation coefficient in S222 is specifically:

[0114]

[0115] The detail coefficients are:

[0116]

[0117] The approximate coefficients and detail coefficients of each layer are further decomposed:

[0118]

[0119] Among them, a j [n] is the approximation coefficient of the jth layer, dj[n] is the detail coefficient of the jth layer; h[k] is the low-pass filter coefficient, and g[k] is the high-pass filter coefficient; is the coefficient of the nth node in the jth layer.

[0120] In one embodiment, the frequency band energy calculation formula in S223 is:

[0121]

[0122] The energy ratio calculation formula is:

[0123]

[0124] Among them, E(f k ) is the energy of the kth frequency band, X(f) is the amplitude at f, and f k1 and f k2 are the starting and ending frequencies of the kth frequency band, and K is the total number of frequency bands.

[0125] Specifically, based on the power signal data collected in this experiment, the db4 wavelet function is selected as the fundamental wave, the signal is decomposed into 6 layers, and a wavelet packet tree is constructed. Each layer of decomposition includes high-frequency and low-frequency filtering of the signal, and then downsampling the filtered signal. The energy proportion of each frequency band is calculated, and 20 effective frequency bands are obtained. The energy change trend of each effective frequency band under different tool wear conditions is analyzed, and the Pearson correlation coefficient method is used to select 7 frequency band energies that are highly correlated with the tool wear state as sample features.

[0126] The features obtained from the time domain, frequency domain and wavelet packet analysis and their corresponding tool state types are encoded to form a sample set, and the tool state sample set is further divided into a training set and a test set. 48 groups in the sample set are selected as training sets and 37 groups as test sets.

[0127] Furthermore, the structure of the BP neural network is as follows Figure 2 As shown in the figure, the training process of BP neural network is as follows:

[0128] Initialize network weights and thresholds: randomly initialize the weights and biases of the BP neural network; input samples and expected outputs: input training data (input samples and expected outputs) into the neural network; calculate actual output and calculate error: calculate the actual output of the neural network through forward propagation, and calculate the output error; adjust weights and thresholds by error back propagation: use the error back propagation algorithm to adjust the weights and biases of the network; meet the accuracy requirements or reach the maximum number of iterations: check whether the preset accuracy requirements are met or the maximum number of iterations is reached; end: if the conditions are met, end the training process.

[0129] Combined with the gray wolf particle swarm algorithm to optimize the BP neural network, such as Figure 3 As shown, in one embodiment, S300 includes:

[0130] S310: input the training set data in the sample set and randomly initialize the weight and bias of the BP neural network, and initialize the position and speed of the particle group and the gray wolf group, and initialize the training parameters; wherein the training parameters include the population size, inertia coefficient, individual learning factor, social learning factor and total number of iterations;

[0131] S320: Calculate the fitness value of each individual according to the fitness function and sort them from high to low, select the first three individuals with the best solutions according to the fitness value, and record their position vectors;

[0132] S330: Update the position and velocity of each particle using the Grey Wolf optimization algorithm;

[0133] S340: Calculate the fitness value of each individual after the update, and compare it with the value of the previous iteration, select the individuals with the top three best positions, and the position vectors of these three optimal solution individuals represent the optimal weights and biases in the current iteration;

[0134] S350: Use the current optimal weights and biases to train the BP neural network, calculate the training error, and continuously adjust the weights and biases until the termination condition is met, thereby completing the optimization of the BP neural network. Further, the termination condition is that the maximum number of iterations is reached or the error meets the requirement.

[0135] In one embodiment, S330 specifically includes:

[0136] The position update formula of each particle is:

[0137] X(t+1)=X(t)+V(t+1)

[0138] The speed update formula is:

[0139] V(t+1)=ωV(t)+c1r1[P b -X(t)]+w α c2r2α [P α -X(t)]+

[0140] w β c2r 2β [P β -X(t)]+w δ c2r 2δ [P δ -X(t)]

[0141] Among them, c1r1[P b -X(t)] is the relationship between the current velocity increment and the particle's best historical position; P α , P β , P δ are the position vectors of the individuals in the optimal solution of SEG, c1 is the individual learning factor, ω is the inertia coefficient, c2 is the social learning factor, r1 and r2 are random numbers between [0,1]; 2α 、r 2β 、r 2δ are the social learning factors corresponding to the three wolves, r 2α 、r 2β 、r 2δ is a random number between [0,1]; w α 、w β and w δ The weights corresponding to the three wolves are set to 0.45, 0.3, and 0.25 respectively.

[0142] Finally, the Pearson correlation coefficient method is used to screen out the features with high correlation and complete the dimensionality reduction of high-dimensional features, such as Figure 4 The characteristic parameters with high correlation with tool wear status are used as the input of BP network, and the tool wear status category is used as the output. The tool status recognition model based on power sensor is trained, and the tool status recognition is performed using the model.

[0143] Specifically, 37 test data sets were input into the trained tool state recognition model, and a tool state recognition accuracy of 95% was obtained. The model can effectively identify the wear state of the tool and provide reliable support for the actual machining process.

[0144] In one embodiment, in S500, the features with correlation coefficients greater than a preset threshold are screened from the feature sample set using the Pearson correlation coefficient method, including:

[0145] S510: Pearson correlation coefficient method is used to calculate the Pearson correlation coefficient between each feature in the sample feature set and the tool wear state, specifically:

[0146]

[0147] Among them, the Pearson correlation r is used to measure the linear correlation between two variables, and its value range is [-1,1], X i and Y i are the i-th sample values ​​of the characteristic variable X and tool wear state Y, and are the means of X and Y respectively;

[0148] S520: Screen out features whose absolute values ​​are greater than a preset threshold and use them as input variables of a subsequent tool wear state identification model.

[0149] Specifically, a correlation threshold is set according to actual needs (the threshold is set to 0.7), which means that a correlation coefficient with an absolute value greater than 0.7 indicates a strong correlation.

[0150] In one embodiment, a tool wear state monitoring system based on a power sensor includes a sample data acquisition module, a sample set construction module, a tool wear state identification model establishment module based on a BP neural network, and a tool wear state judgment module;

[0151] The sample data acquisition module is used to arrange the power sensor at the inverter input end of the machine tool, collect the sample data of the spindle voltage and current values ​​in the machine tool processing process in real time through the sensor, and convert it into a power signal, and record the tool wear status corresponding to each sample data;

[0152] The sample set construction module is used to preprocess the power signal, calculate the time domain characteristics and frequency domain characteristics of the preprocessed power signal, extract the energy characteristics of the frequency band corresponding to tool wear through fast Fourier transform and wavelet packet decomposition, and together form a feature sample set. The feature sample set and the corresponding tool state type are encoded to form a sample set, and divided into a training set and a test set according to a preset ratio;

[0153] The module for establishing a tool wear state identification model based on BP neural network is used to take characteristic parameters as input and tool wear state as output, and to optimize the BP neural network by using the training set and the test set combined with the Grey Wolf Particle Swarm Algorithm to establish a tool wear state identification model based on BP neural network;

[0154] The tool wear state judgment module is used to use the Pearson correlation coefficient method to filter out features with correlation coefficients greater than a preset threshold from the feature sample set, complete the dimensionality reduction of the feature sample set, and use the filtered feature parameters as the input variables of the tool wear state identification model, and the tool wear state as the output variable to determine the mapping relationship between input and output, and judge the tool wear state based on the filtered feature parameters.

[0155] For the specific definition of a tool wear state monitoring system based on a power sensor, please refer to the definition of a tool wear state monitoring method based on a power sensor above, which will not be repeated here. Each module in the above-mentioned tool wear state monitoring system based on a power sensor can be implemented in whole or in part through software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0156] Taking the milling process as an example, the present invention uses a power sensor to collect the power signal in the cutting process; uses time domain analysis, frequency domain analysis and wavelet packet analysis to optimally extract the signal characteristics and frequency band energy of the tool cutting state as sample features to form a sample set; uses the Grey Wolf Particle Swarm Algorithm to optimize the BP neural network and then construct a tool state recognition model to achieve the purpose of automatically identifying the tool wear state.

[0157] The above is a detailed introduction to a tool wear state monitoring method based on a power sensor provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of ​​the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified in a number of ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A tool wear state monitoring method based on a power sensor, characterized in that: The method comprises the following steps: S100: arranging a power sensor at the inverter input end of the machine tool, collecting sample data of the spindle voltage and current values ​​during the machine tool processing in real time through the sensor, and converting them into power signals, while recording the tool wear status corresponding to each sample data; S200: preprocessing the power signal, calculating the time domain features and frequency domain features of the preprocessed power signal, extracting the energy features of the frequency band corresponding to tool wear through fast Fourier transform and wavelet packet decomposition, and jointly forming a feature sample set, encoding the feature sample set and the corresponding tool state type to form a sample set; S300: Taking the characteristic parameters as input and the tool wear status as output, the BP neural network is optimized by using the sample set combined with the Grey Wolf Particle Swarm Optimization Algorithm to establish a tool wear status identification model based on the BP neural network; S400: Use the Pearson correlation coefficient method to filter out features with correlation coefficients greater than a preset threshold from the feature sample set, complete the dimensionality reduction of the feature sample set, and use the filtered feature parameters as input variables of the tool wear state identification model, and the tool wear state as the output variable to determine the mapping relationship between the input and output, and judge the tool wear state based on the filtered feature parameters.

2. The method according to claim 1, characterized in that S100 includes: S110: presetting the tool running trajectory and processing parameters, collecting the current and voltage signals of the spindle at fixed intervals, observing the tool wear state with a microscope, measuring the tool wear amount, and calibrating the tool wear state at this time; S120: Repeat the experiment in this way to obtain sample data of current and voltage values ​​of multiple spindles, and convert them into power signals. At the same time, the tool wear status corresponding to each sample data is recorded.

3. The method according to claim 2, characterized in that The time domain features in S200 include mean, root mean square, standard deviation and peak-to-peak value, and the frequency domain features include frequency mean, frequency centroid and frequency standard deviation. The time domain features and frequency domain features of the preprocessed power signal are calculated in S200, specifically: The mean is: The root mean square is: The standard deviation is: Where X(i) is the measured value of the power signal at the i-th moment, N is the total number of points of the power signal, and x is the average value of the power signal; The peak-to-peak value is: PP=max(x(i))-min(x(i)) Wherein, max(x(i)) is the maximum value of the power signal data, and min(x(i)) is the minimum value of the power signal; The frequency mean is: The frequency center of gravity is specifically: The frequency standard deviation is: Where k is the total number of spectral lines, f(k) is the frequency value of the kth spectral line, and |X(k)| is the absolute value of the power spectrum value corresponding to the kth spectral line.

4. The method according to claim 3, characterized in that In S200, fast Fourier transform and wavelet packet decomposition are used to extract the energy characteristics of the frequency band corresponding to tool wear, which together form a feature sample set, specifically: S210: Performing fast Fourier transform on the power signal after noise reduction to convert the time domain signal into a frequency domain signal; S220: Perform wavelet packet decomposition on the power signal after noise reduction using wavelet basis functions, decompose the signal into frequency bands, calculate the proportion of each frequency band energy to the total energy, and select the frequency band whose energy proportion exceeds a preset threshold as the effective frequency band; by analyzing the change trend of the energy of each effective frequency band under different tool wear conditions, select the frequency band energy that is highly correlated with the tool wear condition as the sample feature.

5. The method according to claim 4, characterized in that The specific steps of wavelet packet decomposition in S220 are as follows: S221: performing multi-layer decomposition on the power signal after noise reduction using wavelet basis functions to construct a wavelet packet tree; each layer of decomposition includes high-frequency and low-frequency filtering on the power signal after noise reduction, and then down-sampling the filtered signal; S222: The power signal x(t) after noise reduction is passed through a low-pass filter h(t) and a high-pass filter g(t) to obtain approximate coefficients and detail coefficients respectively, and the approximate coefficients and detail coefficients are further subjected to low-pass and high-pass filtering to obtain coefficients of the next layer, and this process is repeated until a predetermined number of decomposition layers is reached; S223: Calculate the square sum of the approximate coefficient and the detail coefficient of each frequency band to obtain the energy value of each frequency band, calculate the proportion of the energy of each frequency band to the total energy, and select the frequency band whose energy proportion exceeds a preset threshold as the effective frequency band.

6. The method according to claim 5, characterized in that The approximate coefficients in S222 are: The detail coefficients are: The approximate coefficients and detail coefficients of each layer are further decomposed: Among them, a j [n] is the approximation coefficient of the jth layer, dj[n] is the detail coefficient of the jth layer; h[k] is the low-pass filter coefficient, and g[k] is the high-pass filter coefficient; is the coefficient of the nth node in the jth layer.

7. The method according to claim 6, characterized in that The formula for calculating the energy in the S223 mid-band is: The energy ratio calculation formula is: Among them, E(f k ) is the energy of the kth frequency band, f k1 and f k2 are the starting and ending frequencies of the kth frequency band, and K is the total number of frequency bands.

8. The method according to claim 7, characterized in that S300 includes: S310: Input the sample set and randomly initialize the weight and bias of the BP neural network, initialize the position and speed of the particle swarm and the wolf pack, and initialize the training parameters; wherein the training parameters include the population size, inertia coefficient, individual learning factor, social learning factor and total number of iterations; S320: Calculate the fitness value of each individual according to the fitness function and sort them from high to low, select the first three individuals with the best solutions according to the fitness value, and record their position vectors; S330: Update the position and velocity of each particle using the Grey Wolf optimization algorithm; S340: Calculate the fitness value of each individual after the update, and compare it with the value of the previous iteration, select the individuals with the top three best positions, and the position vectors of these three optimal solution individuals represent the optimal weights and biases in the current iteration; S350: Use the current optimal weights and biases to train the BP neural network, calculate the training error, and continuously adjust the weights and biases until the termination condition is met, thereby completing the optimization of the BP neural network.

9. The method according to claim 8, characterized in that S330 is specifically: The position update formula of each particle is: X(t+1)=X(t)+V(t+1) The speed update formula is: V(t+1)=ωV(t)+c1r1[P b -X(t)]+w α c2r 2α [P α -X(t)]+w β c2r 2β [P β -X(t)]+w δ c2r 2δ [P δ -X(t)] Among them, c1r1[P b -X(t)] is the relationship between the current velocity increment and the particle's best historical position; P α , P β , P δ are the position vectors of the individuals in the optimal solution of SEG, c1 is the individual learning factor, ω is the inertia coefficient, c2 is the social learning factor, r1 and r2 are random numbers between [0, 1]; 2α 、r 2β 、r 2δ are the social learning factors corresponding to the three wolves, r 2α 、r 2β 、r 2δ is a random number between [0, 1]; w α 、w β and w δ The weights corresponding to the three wolves are set to 0.45, 0.3, and 0.25 respectively.

10. The method according to claim 9, characterized in that In S500, the Pearson correlation coefficient method is used to filter out features with correlation coefficients greater than a preset threshold from the feature sample set, including: S510: Pearson correlation coefficient method is used to calculate the Pearson correlation coefficient between each feature in the sample feature set and the tool wear state, specifically: Among them, the Pearson correlation r is used to measure the linear correlation between two variables, and its value range is [-1, 1]. i and Y i are the i-th sample values ​​of the characteristic variable X and tool wear state Y, and are the means of X and Y respectively; S520: Screen out features whose absolute values ​​are greater than a preset threshold and use them as input variables of a subsequent tool wear state identification model.

Images