Cutter wear state identification method and system for cutting of numerical control machine tool

Abstract

The invention discloses a tool wear state identification method and system for cutting of a numerical control machine tool. The method comprises the following steps: acquiring a three-way vibration signal, a power signal and a temperature signal; performing time domain, frequency domain and time-frequency domain multi-domain feature extraction on the three-way vibration signal, the power signal and the temperature signal, and constructing a multi-domain high-dimensional feature vector; performing feature fusion dimension reduction on the high-dimensional feature vector to obtain a feature sample set; obtaining a cutter state recognition model according to the feature sample set; and inputting the acquired multi-source signals into a cutter state identification model to carry out cutter wear state identification of cutting machining of the numerical control machine tool. By the adoption of the technical scheme, various machining conditions and tool types can be better adapted, and the applicability and universality of the recognition system are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of mechanical manufacturing and processing, and in particular relates to a method and system for identifying tool wear status in cutting processing of numerically controlled machine tools. Background Art

[0002] In the cutting process of CNC machine tools, tools play a vital role. As a component that directly contacts the workpiece material, the performance and state of the tool directly determine the machining accuracy and the surface quality of the final workpiece. The working process of the tool involves high-speed rotation or movement, and material removal under high temperature and high pressure conditions, which makes it inevitable that the tool will wear during use. Changes in the wear state not only affect the machining quality, but may also cause damage to the tool and even cause the machining process to fail. Therefore, accurately identifying the wear state of the tool is the key to ensuring the stability of the machining process and improving production efficiency.

[0003] At present, the methods of tool wear state identification are divided into two categories: direct method and indirect method. The direct method determines the state of the tool by directly measuring the size of the tool wear surface or the change of the cutting edge shape when the tool is damaged, such as optical method, radiation method, resistance method, computer vision method, etc. Its advantage is high accuracy, but it is difficult to achieve online measurement. The indirect method extracts the feature information corresponding to the feature by identifying one or several parameters that have a strong intrinsic connection with tool wear and damage, such as: cutting force signal, vibration signal, acoustic emission signal, motor power or current signal, cutting temperature, workpiece surface roughness, etc., and establishes the corresponding relationship between these feature signals and the processing tool state, and uses pattern recognition methods for classification and identification. The indirect method does not affect the cutting process and can complete online real-time identification and monitoring, but there is still a problem of insufficient generalization ability of the identification method. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for identifying tool wear status in CNC machine tool cutting processing, which not only realizes efficient and accurate tool wear status identification, but also overcomes the insufficient generalization ability of traditional technologies; in particular, it shows significant advantages in online monitoring, real-time processing, model optimization, etc., and has broad application prospects.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] A method for identifying tool wear status in CNC machine tool cutting, comprising:

[0007] Step S1, collecting three-way vibration signals, power signals, and temperature signals;

[0008] Step S2, extracting multi-domain features in the time domain, frequency domain, and time-frequency domain of the three-way vibration signal, the power signal, and the temperature signal, and constructing a multi-domain high-dimensional feature vector;

[0009] Step S3, performing feature fusion and dimensionality reduction on the high-dimensional feature vector to obtain a feature sample set;

[0010] Step S4, obtaining a tool state recognition model according to the feature sample set;

[0011] Step S5: input the collected multi-source signals into a tool state recognition model to perform tool wear state recognition in CNC machine tool cutting.

[0012] Preferably, in step S2, the maximum value, maximum absolute value, minimum value, mean value, peak-to-peak value, absolute average value, root mean square value, standard deviation, kurtosis, skewness, margin index, waveform index, pulse index, and peak index corresponding to the three-way vibration signal, power signal, and temperature signal are extracted as characteristic parameters; the frequency domain is used to extract the center of gravity frequency, average frequency, frequency root mean square, and frequency variance of the three-way vibration signal as characteristic parameters; the three-way vibration signal is decomposed by a three-layer wavelet packet, and the energy proportion of each decomposition frequency band is extracted, and a total of 124 characteristic parameters are obtained.

[0013] Preferably, in step S3, a kernel principal component analysis algorithm is used to perform feature fusion and dimensionality reduction on 124 feature parameters, and the number of principal components is determined according to the principal component variance cumulative contribution percentage method to obtain a 13-dimensional feature sample set.

[0014] Preferably, in step S4, a model combining a sparrow search algorithm and a BP neural network is trained according to the input of the feature sample set to obtain a tool state recognition model.

[0015] The present invention also provides a tool wear state identification system for CNC machine tool cutting, comprising:

[0016] A collection device, used for collecting three-way vibration signals, power signals, and temperature signals;

[0017] An extraction device is used to perform multi-domain feature extraction on the three-way vibration signal, the power signal and the temperature signal in the time domain, the frequency domain and the time-frequency domain to construct a multi-domain high-dimensional feature vector;

[0018] A processing device, used for performing feature fusion and dimensionality reduction on high-dimensional feature vectors to obtain a feature sample set;

[0019] A training device, used for obtaining a tool state recognition model based on a feature sample set;

[0020] The identification device is used to input the collected multi-source signals into the tool state identification model to perform tool wear state identification in CNC machine tool cutting processing.

[0021] Preferably, the extraction device is used to extract the maximum value, maximum absolute value, minimum value, mean value, peak-to-peak value, absolute average value, root mean square value, standard deviation, kurtosis, skewness, margin index, waveform index, pulse index, and peak index corresponding to the three-way vibration signal, power signal, and temperature signal as characteristic parameters; perform frequency domain extraction of the center of gravity frequency, average frequency, frequency root mean square, and frequency variance of the three-way vibration signal as characteristic parameters; perform three-layer wavelet packet decomposition on the three-way vibration signal to extract the energy proportion of each decomposition frequency band, and obtain a total of 124 characteristic parameters.

[0022] Preferably, the processing device uses a kernel principal component analysis algorithm to perform feature fusion and dimensionality reduction on 124 feature parameters, determines the number of principal components according to the principal component variance cumulative contribution percentage method, and obtains a 13-dimensional feature sample set.

[0023] Preferably, the training device will train a model combining a sparrow search algorithm and a BP neural network based on the input of the feature sample set to obtain a tool state recognition model.

[0024] The present invention constructs a multi-domain high-dimensional feature vector, uses the kernel principal component analysis algorithm to perform feature fusion and dimensionality reduction on the high-dimensional feature vector to solve the problem of information redundancy in the high-dimensional feature vector, and constructs a model combining the sparrow search algorithm and the BP neural network to identify the tool wear state. Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. The present invention can fully capture the key parameters in cutting processing. This multi-dimensional data collection method ensures a comprehensive understanding of the tool wear state and helps to improve the accuracy of identification.

[0026] 2. The present invention uses the kernel principal component analysis algorithm to reduce the dimensionality of high-dimensional features, effectively integrating and compressing feature information. This process not only reduces the data dimension and reduces the computational burden, but also retains the most important feature information and improves the processing efficiency of the model.

[0027] 3. The present invention uses a model combining a sparrow search algorithm with a BP neural network to identify tool status, combining adaptive and optimization capabilities. This model performs well in dealing with nonlinear problems and can improve the accuracy and robustness of tool wear status identification.

[0028] 4. The comprehensive utilization of multiple signals and features of the present invention enables the solution to better adapt to various processing conditions and tool types, thereby improving the applicability and versatility of the recognition system. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0030] Figure 1 It is a flowchart of a method for identifying tool wear status in CNC machine tool cutting processing according to an embodiment of the present invention;

[0031] Figure 2 This is a diagram showing the contribution rate of dimensionality reduction of a high-dimensional feature vector according to an embodiment of the present invention;

[0032] Figure 3 It is a tool state recognition distribution diagram combining the sparrow search algorithm and the BP neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] Embodiment 1:

[0036] The embodiment of the present invention provides a method for identifying tool wear status in CNC machine tool cutting. First, multi-source signals are collected by vibration collection equipment, power collection equipment, and temperature collection equipment. Secondly, multi-domain feature extraction is performed on the multi-source signals to construct a high-dimensional feature vector. Thirdly, a representative feature vector is constructed using the kernel principal component analysis method. Finally, a model combining the sparrow search algorithm and the BP neural network is used to identify the tool status, such as Figure 1 As shown, the following steps are included:

[0037] Step 1: Collect the three-way vibration signal in the cutting process through the three-way vibration signal acquisition device, collect the power signal in the cutting process through the power signal acquisition device, and collect the temperature signal in the temperature cutting process through the temperature signal acquisition device. (The three-way vibration signal acquisition device, power signal acquisition device, and temperature signal acquisition device need to set sampling parameters: such as sampling time, sampling frequency, and number of sampling points).

[0038] Step 2: Divide the tool wear state into three stages according to the needs of processing: initial wear, normal wear, and rapid wear. In order to obtain these three tools with different degrees of wear, select the largest cutting parameters as much as possible according to the performance of the machine tool and the characteristics of the tool and the material; process 600 sets of original data. In the CNC machining process, the vibration signal collected by the three-way vibration sensor not only truly reflects the vibration state signal, but also inevitably mixed with noise signals. In view of the limitations of soft and hard threshold functions in processing signal denoising, the adjustment parameter L and exponential function are introduced to give the new threshold function adaptability and continuity, respectively, thereby improving the denoising effect. The expression of the improved threshold function is as follows:

[0039]

[0040] Among them, w j,k represents the original wavelet coefficients; represents the wavelet coefficient after threshold processing; sign() represents the step function; λ represents the threshold. L is an adjustment parameter, and the value of L determines the characteristics of the threshold function. When L = 0, the function is equivalent to the soft threshold function; when L → +∞, the function is a hard threshold function; when L takes a moderate value, it can effectively make up for the shortcomings of the soft and hard threshold functions and optimize the denoising effect.

[0041] Then, the maximum value, maximum absolute value, minimum value, mean value, peak-to-peak value, absolute average value, root mean square value, standard deviation, kurtosis, skewness, margin index, waveform index, pulse index, and peak index corresponding to the three-way vibration signal, power signal, and temperature signal are extracted as feature parameters (14 for each channel signal, a total of 70 feature parameters); the frequency domain is used to extract the center of gravity frequency, average frequency, frequency root mean square, and frequency variance of the three-way vibration signal as feature parameters (4 for each of the three-way vibration signals, a total of 12 feature parameters); the three-way vibration signal is decomposed by 3-layer wavelet packets to extract the energy proportion of each decomposition frequency band (14 for each of the three-way vibration signals, a total of 42 feature parameters); a total of 124 feature parameters are obtained, and the features are concatenated end to end to construct a time domain-frequency domain-time-frequency domain joint feature vector.

[0042] Step 3: First, perform kernel principal component analysis based on the joint eigenvector constructed in step 2, and use the kernel function Project the joint feature matrix T into a linearly separable high-dimensional space for kernel principal component analysis, and obtain the principal component coefficient matrix Coeff as follows:

[0043]

[0044] The eigenvalue λ1≥λ2≥···≥λ d and the corresponding eigenvector α k(k=1,2,···,d). Then the principal component variance cumulative contribution rate percentage method is used to determine the number of principal components and the ratio It reflects the contribution of the kth component in the generated feature space to the overall variance. The ratio corresponding to the important component is relatively large. The system dimension can be reduced by retaining the first s eigenvectors. The number of principal components s is generally determined according to the following formula (3):

[0045]

[0046] In this case, according to formula (3), the main component with a contribution rate of 95% was selected. After calculation and analysis, 13 main components were finally determined. These main components captured most of the important information in the data, such as Figure 2 As shown. By projecting the principal component coefficient matrix of the constructed joint eigenvector, the fused eigenvector after dimensionality reduction can be obtained. Therefore, a 13-dimensional feature sample set is obtained;

[0047] Step 4: First, divide the 600 groups of samples obtained after feature fusion and dimensionality reduction into training samples (480 groups, i.e. 4 / 5) and test samples (120 groups, i.e. 1 / 5), and normalize the data to [0,1]; then initialize the parameters of the sparrow search algorithm (SSA), including the size of the sparrow population, the maximum number of iterations, the proportion of discoverers, the proportion of sparrows that are aware of danger, and the number of discoverers; the BP neural network includes an input layer, a hidden layer, and an output layer. Determine the number of hidden layers, the number of input layer neurons, the number of hidden layer neurons, and the number of output layer neurons, where the input layer is the 13 principal components obtained in step 3, so the number of input layer neurons is 13, and the output layer is the category of three tool states, so the number of output layer neurons is 3; the number of hidden layer neurons is determined by empirical methods; in this embodiment, the number of hidden layer neurons is 7, and the number of training times is 1000 times. Substitute the normalized training set data into the BP neural network, use the SSA algorithm to train the BP neural network, and form a model based on the SSA-BP neural network. The optimal position and the best fitness value are output by the SSA algorithm and assigned to the BP neural network as parameter weights and thresholds, respectively. The initial sparrow population size is 30, the maximum number of iterations is 100, the bird flock warning value is 0.8, the proportion of discoverers is 0.2, and the proportion of vigilants is 0.3.

[0048] The finder's position update formula is as follows:

[0049]

[0050] Among them, T max is the maximum number of iterations; t is the current iteration; j = 1, 2, ..., d; X i,j(t) represents the next value of the jth sparrow at the tth iteration; α is a random number between 0 and 1, R2 (alarm value) is a number in the range of 0 to 1; ST (safety threshold) is between 0.5 and 1; Q is a random number based on normal distribution; L represents a 1×d matrix, where each element is 1. If R2≥ST, some sparrows have discovered the hunter, and all sparrows must quickly fly to other safe areas; when R2<ST, it means that there is no hunter around, and the discoverer enters a wide search mode.

[0051] Since the follower mainly relies on the information of the discoverer to find food resources, its position update strategy may be more conservative in order to follow the discoverer more stably. The position update of the follower is defined according to the following formula.

[0052]

[0053] Among them, X p is the best position for the discoverer to occupy; X worst represents the current global worst position; A represents a 1×d matrix, which randomly assigns 1 or -1 to each element. + =A T (AA T ) -1 ; If i>n / 2, this indicates that followers with poorer fitness values ​​are more likely to starve.

[0054] After updating the position of the entire population, some sparrows are selected as sentinels to identify and warn of danger. They usually account for 10% to 20% of the population, and their initial positions are randomly generated. The position update of the sentinels is also conservative, as shown in the following formula:

[0055]

[0056] Among them, X best is the current global optimal position; ε is a small constant to avoid zero division error; is a random number that follows a standard normal distribution; f g and f w are the current global best and worst fitness values ​​respectively; k is a random number between 1 and -1; f i is the current sparrow’s fitness value; f i =f g , indicating that the sparrows in the middle of the population are aware of the danger and should move closer to other sparrows. i >f g , then the sparrow is on the edge of the population.

[0057] The optimal position and the best fitness value are outputted as above, and are assigned to the BP neural network as parameter weights and thresholds, respectively, so as to obtain the optimal SSA-BP neural network model; then the test sample is input into the tool state recognition model to identify the tool wear state. Figure 3 As shown in the figure, the test samples (120 groups, 40 groups of data for each tool state) are input, "·" represents the real label, and "×" represents the predicted label. According to the results, only 3 groups of prediction errors are made in the 120 groups of data. Therefore, the tool recognition accuracy based on the SSA-BP neural network model is improved compared with the BP neural network model, and the accuracy of the SSA-BP neural network model reaches 97.5%.

[0058] Embodiment 2:

[0059] The present invention also provides a tool wear state identification system for CNC machine tool cutting processing, comprising:

[0060] A collection device, used to collect three-way vibration signals, power signals, and temperature signals;

[0061] An extraction device is used to perform multi-domain feature extraction on the three-way vibration signal, the power signal and the temperature signal in the time domain, the frequency domain and the time-frequency domain to construct a multi-domain high-dimensional feature vector;

[0062] A processing device, used for performing feature fusion and dimensionality reduction on high-dimensional feature vectors to obtain a feature sample set;

[0063] A training device, used for obtaining a tool state recognition model based on a feature sample set;

[0064] The identification device is used to input the collected multi-source signals into the tool state identification model to perform tool wear state identification in CNC machine tool cutting processing.

[0065] As an implementation mode of an embodiment of the present invention, the extraction device is used to extract the maximum value, maximum absolute value, minimum value, mean value, peak-to-peak value, absolute average value, root mean square value, standard deviation, kurtosis, skewness, margin index, waveform index, pulse index, and peak index corresponding to the three-way vibration signal, power signal, and temperature signal as characteristic parameters; perform frequency domain extraction of the center of gravity frequency, average frequency, frequency root mean square, and frequency variance of the three-way vibration signal as characteristic parameters; perform three-layer wavelet packet decomposition on the three-way vibration signal to extract the energy proportion of each decomposition frequency band, and obtain a total of 124 characteristic parameters.

[0066] As an implementation method of an embodiment of the present invention, the processing device uses a kernel principal component analysis algorithm to perform feature fusion and dimensionality reduction on 124 feature parameters, determines the number of principal components according to the principal component variance cumulative contribution rate percentage method, and obtains a 13-dimensional feature sample set.

[0067] As an implementation method of the present invention, the training device will train the model combined with the sparrow search algorithm and the BP neural network according to the feature sample set input to obtain the tool state recognition model.

[0068] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for identifying tool wear status in CNC machine tool cutting, characterized in that: include: Step S1, collecting three-way vibration signals, power signals, and temperature signals; Step S2, extracting multi-domain features in the time domain, frequency domain, and time-frequency domain of the three-way vibration signal, the power signal, and the temperature signal, and constructing a multi-domain high-dimensional feature vector; Step S3, performing feature fusion and dimensionality reduction on the high-dimensional feature vector to obtain a feature sample set; Step S4, obtaining a tool state recognition model according to the feature sample set; Step S5: input the collected multi-source signals into a tool state recognition model to perform tool wear state recognition in CNC machine tool cutting.

2. The method for identifying tool wear status in CNC machine tool cutting processing according to claim 1, characterized in that: In step S2, the maximum value, maximum absolute value, minimum value, mean value, peak-to-peak value, absolute average value, root mean square value, standard deviation, kurtosis, skewness, margin index, waveform index, pulse index, and peak index corresponding to the three-way vibration signal, power signal, and temperature signal are extracted as characteristic parameters; the frequency domain is used to extract the center of gravity frequency, average frequency, frequency root mean square, and frequency variance of the three-way vibration signal as characteristic parameters; the three-way vibration signal is decomposed by a three-layer wavelet packet to extract the energy proportion of each decomposition frequency band, and a total of 124 characteristic parameters are obtained.

3. The method for identifying tool wear status in CNC machine tool cutting processing according to claim 2, characterized in that: In step S3, the kernel principal component analysis algorithm is used to perform feature fusion and dimensionality reduction on the 124 feature parameters, and the number of principal components is determined according to the principal component variance cumulative contribution rate percentage method to obtain a 13-dimensional feature sample set.

4. The method for identifying tool wear status in CNC machine tool cutting processing according to claim 3, characterized in that: In step S4, a model combining a sparrow search algorithm and a BP neural network is trained based on the input of the feature sample set to obtain a tool state recognition model.

5. A tool wear state identification system for CNC machine tool cutting, characterized in that: include: A collection device, used to collect three-way vibration signals, power signals, and temperature signals; An extraction device is used to perform multi-domain feature extraction on the three-way vibration signal, the power signal and the temperature signal in the time domain, the frequency domain and the time-frequency domain to construct a multi-domain high-dimensional feature vector; A processing device, used for performing feature fusion and dimensionality reduction on high-dimensional feature vectors to obtain a feature sample set; A training device, used for obtaining a tool state recognition model based on a feature sample set; The identification device is used to input the collected multi-source signals into the tool state identification model to perform tool wear state identification in CNC machine tool cutting processing.

6. The tool wear state identification system for CNC machine tool cutting processing according to claim 5, characterized in that: The extraction device is used to extract the maximum value, maximum absolute value, minimum value, mean value, peak-to-peak value, absolute average value, root mean square value, standard deviation, kurtosis, skewness, margin index, waveform index, pulse index, and peak index corresponding to the three-dimensional vibration signal, power signal, and temperature signal as characteristic parameters; perform frequency domain extraction of the three-dimensional vibration signal to obtain the center of gravity frequency, average frequency, frequency root mean square, and frequency variance as characteristic parameters; perform three-layer wavelet packet decomposition on the three-dimensional vibration signal to extract the energy proportion of each decomposition frequency band, and obtain a total of 124 characteristic parameters.

7. The tool wear state identification system for CNC machine tool cutting processing according to claim 6, characterized in that: The processing device uses the kernel principal component analysis algorithm to perform feature fusion and dimensionality reduction on 124 feature parameters, and determines the number of principal components according to the principal component variance cumulative contribution rate percentage method to obtain a 13-dimensional feature sample set.

8. The method for identifying tool wear status in CNC machine tool cutting processing according to claim 7, characterized in that: The training device will train the model combined with the sparrow search algorithm and the BP neural network according to the feature sample set input to obtain the tool state recognition model.

Images