Evolutionary clustering analysis based tool wear modeling method

Abstract

In order to solve the technical problems that the tool wear model obtained by the existing tool wear modeling method is inaccurate when fitting the actual tool wear trend, and the existing cutting tool wear monitoring method is only limited to statistical analysis of the tool wear evolution process and is not conducive to actual production processing, the present application provides a tool wear modeling method based on evolutionary clustering analysis, which sequentially performs consistency, stage and correlation analysis on the evolutionary clustering distribution diagram, and on the basis of the analysis, a tool wear model with high fitting accuracy to the actual tool wear trend is constructed, and the tool wear evolution law can be reused by using the tool wear model established by the present application in new data.

Description

Technical Field

[0001] This invention relates to a tool wear modeling method, and more particularly to a tool wear modeling method based on evolutionary clustering analysis. Background Technology

[0002] Tool wear is one of the major problems in machining and manufacturing, directly affecting the surface quality of cutting tools, production efficiency, and personal safety. In modern manufacturing, downtime caused by tool damage accounts for 20% of total downtime, and tool use and replacement costs account for 3%-12% of total machining costs. Analyzing the tool wear process using appropriate methods, accurately describing the changing trends of tool wear, and establishing a tool wear model that can accurately fit the actual tool wear curve is an urgent problem to be solved in machining.

[0003] During machining, under the combined action of the cutting tool, workpiece, machine tool, and cutting fluid, the cutting tool is simultaneously affected by physical and chemical factors such as force, heat, and fatigue. The degree of tool wear increases irreversibly with the progress of cutting, and this increase follows a certain regularity. To mathematically describe the regularity of the tool wear process, existing research in the field mainly focuses on establishing tool wear models based on wear mechanisms, field experience, or mathematical derivations. However, the accuracy of existing tool wear models in fitting actual tool wear trends remains relatively low.

[0004] To address the inaccuracy of existing tool wear modeling methods in fitting actual tool wear trends, patent CN 110647943 B proposes a cutting tool wear monitoring method based on evolutionary data clustering analysis. This method focuses on two types of evolutionary data during tool wear: sensor signals and tool wear values. Through sensor signal feature extraction, feature selection, feature matrix construction and normalization, and evolutionary clustering analysis, a dynamic time-series evolutionary clustering distribution map is obtained, revealing the evolutionary processes of sensor signals and tool wear states, as well as the correlation between them. However, the evolutionary clustering distribution map obtained by patent CN 110647943B is limited to statistical analysis of the tool wear evolution process and does not mathematically summarize the tool wear process, making it difficult to apply the experimentally obtained tool wear patterns to actual production and processing data. Summary of the Invention

[0005] To address the inaccuracies in fitting actual tool wear trends by existing tool wear modeling methods, and the limitations of existing cutting tool wear monitoring methods which are confined to statistical analysis of tool wear evolution and thus unsuitable for practical production processing, this invention proposes a tool wear modeling method based on evolutionary clustering analysis.

[0006] The technical solution of this invention is:

[0007] Tool wear modeling methods based on evolutionary clustering analysis, including

[0008] Step 1: Obtain a series of time-series cluster distribution maps of tool wear evolution;

[0009] Step 1.1 Collect raw evolutionary data;

[0010] Step 1.2: Extract features from the sensor signals;

[0011] Step 1.3 Select sensor signal characteristics that are sensitive to but not redundant to changes in the average width VB of the wear band;

[0012] Step 1.4 Construct a feature matrix using the sensor signal features selected in Step 1.3 and normalize it;

[0013] Step 1.5 involves performing evolutionary cluster analysis on the feature matrix obtained in Step 1.4 to obtain a series of time-series tool wear evolution cluster distribution maps;

[0014] Its characteristic is that it further includes the following steps:

[0015] Step 2: For each data point in the tool wear evolution clustering distribution map, obtain its standard cluster and the number of deviations from its corresponding standard cluster, and then calculate the deviation coefficient EC of the evolution clustering distribution map. If EC is greater than the set value, the consistency is considered good. At this time, the standard clustering distribution map is constructed from the standard clusters of each data point, and proceed to step 3; otherwise, the consistency is considered poor, and return to step 1.3.

[0016] Step 3: Based on the stage-specific characteristics of the standard cluster distribution map, divide it into multiple distinct continuous intervals to reflect different stages of the tool wear evolution process, and calculate the cluster density ρ of each continuous interval. s～e ; s and e are the starting and ending points of the data in the continuous interval s to e, respectively;

[0017] Step 4: For the multiple different continuous intervals, calculate the average feature rate of the normalized feature matrix obtained from the sensor signal. and average wear rate Perform a correlation analysis with the cluster density. If the correlation meets the requirements, proceed to step 5; otherwise, return to step 1.3.

[0018] Step 5: Construct the tool wear model:

[0019] Step 5.1 Construct basic constraints:

[0020]

[0021] In the formula, VB is the average width of the wear band; ΔVB is the differential form of VB; t is the cutting time; Δt is the differential form of t; t0 is the approximation value of t; w(t) is the tool wear model, which satisfies VB=w(t); w′(t) is the first derivative of w(t); w″(t) is the second derivative of w(t);

[0022] Step 5.2: Construct the tool wear model w(t):

[0023] w(t) = w 1 (t)+w 2 (t)+......+w k (t)

[0024] In the formula, w 1 (t), w 2 (t), ..., w k (t) represent the sub-equations corresponding to different continuous intervals, and each sub-equation is based on the cluster density and average characteristic rate of the different continuous intervals. The trend charts of cluster density and average tool wear rate in different continuous intervals are determined by combining the slope variation characteristics of various basic functions.

[0025] Each sub-equation satisfies the basic constraints constructed in step 5.1, and the requirements are: a) different sub-equations control different evolution stages; b) the slope of different sub-equations in other evolution stages is less than the slope in their respective evolution stages.

[0026] Furthermore, the specific method for obtaining the deviation coefficient EC in step 2 is as follows:

[0027] Step 2.1 Count the number of deviations of each data point from the standard cluster;

[0028] Step 2.1.1 Calculate the total number of occurrences of all data points in the cluster distribution map of tool wear evolution over a statistical time series. The mode of each data point in the time series, belonging to its respective cluster. For any data point, the cluster with the most memberships is taken as the standard cluster for that data point;

[0029] Step 2.1.2 For each data point, count the number of deviations that data point makes from its corresponding standard cluster.

[0030]

[0031] Step 2.2 Calculate the deviation coefficient EC:

[0032]

[0033] In the formula, This is the sum of the number of deviations for all data points; This represents the total number of occurrences of all data points.

[0034] Furthermore, the cluster density of each consecutive interval in step 3 is calculated according to the following formula:

[0035]

[0036] In the formula, O e and o s These represent the cluster categories to which the starting point and ending point of the data in the continuous interval s to e belong, respectively.

[0037] Furthermore, the average wear rate of each continuous interval in step 4 Calculate according to the following formula:

[0038]

[0039] The average characteristic rate is calculated according to the following formula:

[0040]

[0041] Furthermore, in step 4, the average feature rate of the normalized feature matrix obtained from the sensor signal... and average wear rate The method for performing correlation analysis with the cluster density is as follows:

[0042] For each continuous interval, calculate the average characteristic rate. The Pearson correlation coefficient and mutual information coefficient with the cluster density ρs~e, and the calculation of the average wear rate. Pearson correlation coefficient and mutual information coefficient with cluster density ρs~e;

[0043] When the Pearson correlation coefficient and mutual information coefficient corresponding to all continuous intervals are greater than the set threshold, it indicates that the correlation meets the requirements. At this time, it is considered that the different continuous intervals divided in step 3 are consistent with the changing trend of the evolution data in the processing process.

[0044] The present invention also provides a non-volatile computer-readable storage medium storing a computer program thereon, wherein the computer program, when run, is used for the above-mentioned tool wear modeling method based on evolutionary clustering analysis.

[0045] The present invention also provides an electronic device, including a processor and a storage medium, wherein a computer program is stored on the storage medium, and the computer program is characterized in that: when the processor runs the computer program, it is used to execute the above-described tool wear modeling method based on evolutionary clustering analysis.

[0046] The beneficial effects of this invention are:

[0047] This invention overcomes the problem that existing evolutionary clustering analysis methods cannot easily apply tool wear evolution clustering distribution maps to actual production and processing data. By performing a series of processing steps on the tool wear evolution clustering distribution map, a tool wear model with high fitting accuracy to the actual tool wear trend is established. The tool wear evolution law can be reused in new data by using the tool wear model established by this invention. Attached Figure Description

[0048] Figure 1 This is the overall flowchart of the tool wear modeling method of the present invention.

[0049] Figure 2 This is a schematic diagram illustrating the principle of acquiring sensor signals during the cutting process using a force sensor and a vibration sensor, according to an embodiment of the present invention.

[0050] Figure 3 This is the tool wear curve obtained through cutting experiments in an embodiment of the present invention.

[0051] Figure 4 This is a schematic diagram of the cutting force signals in three directions collected in an embodiment of the present invention.

[0052] Figure 5 This is a schematic diagram of the vibration signals collected in an embodiment of the present invention.

[0053] Figure 6 This is a trend chart showing the maximum value of the cutting force signal in the X direction collected in an embodiment of the present invention.

[0054] Figure 7 This is a partial example of a series of time-series tool wear evolution cluster distribution maps obtained in the embodiments of the present invention (the number of cutting operations in the figure corresponds to the total number of data points, and the number of clusters corresponds to the number of data point shapes).

[0055] Figure 8 This is a distribution diagram of the standard clusters obtained in an embodiment of the present invention.

[0056] Figure 9 The average feature rate and cluster density obtained in the embodiments of the present invention are given.

[0057] Figure 10 The average wear rate and cluster density obtained in the embodiments of the present invention are shown.

[0058] Figure 11 This is the result of fitting the tool wear curve to the tool wear model obtained in the embodiments of the present invention.

[0059] Figure 12 The result is the fit of the tool wear curve to the existing tool wear model 2.

[0060] Figure 13 The result is the fit of the tool wear curve to the existing tool wear model 3.

[0061] Figure 14 The result is the fit of the tool wear curve to the existing tool wear model 4. Detailed Implementation

[0062] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0063] In this embodiment, a five-axis CNC milling machine is selected as the machining tool for the workpiece, a single-tooth insert milling cutter is selected as the milling cutter, the workpiece material is GH4169 high-temperature alloy, and the dimensions are 196mm × 120mm × 16mm. Force sensors and vibration sensors are used to collect signals closely related to tool wear during workpiece machining. The specific tool wear modeling method is as follows:

[0064] Step 1: Obtain a series of time-series cluster distribution maps of tool wear evolution.

[0065] Step 1.1 Collect raw evolutionary data;

[0066] Reference Figure 2 The force measuring table is installed on the worktable of a five-axis CNC milling machine. A fixture consisting of a pressure plate and pads is used to fix the workpiece to the force measuring table. A vibration sensor is installed on the side wall of the workpiece. Tool setting is performed with the lower left corner of the workpiece as the machining origin. The machining path is along the length of the workpiece from left to right, and the machining length for each pass is the workpiece length. Under the same machining parameters, the milling cutter is removed after each certain cutting distance or time. The average width VB of the wear band on the flank face of the milling cutter (or tool wear value) is measured using an Alicona fully automatic tool measuring instrument. The average width VB of the wear band is typical evolutionary data, increasing irreversibly with increasing cutting distance at different rates. One average width VB of the wear band corresponds to a set of acquired original force signals and a set of original vibration signals. Each set of original evolutionary data acquired in this step includes the average width VB of the wear band, force signals, and vibration signals. The tool wear curve obtained from the cutting experiment is shown below. Figure 3 As shown, the cutting force signal samples collected in the three directions are as follows: Figure 4 As shown, the collected vibration signal samples are as follows: Figure 5 As shown.

[0067] Step 1.2: Extract features from the sensor signals;

[0068] Because the original signal has a very large data volume and high data dimensionality, coupled with a lot of redundancy and interference information, the computational load of the subsequent evolutionary clustering model increases. To reduce the data volume and dimensionality of the original signal, this invention uses time-domain, frequency-domain, and time-frequency-domain analysis methods to extract features from the original signal. Table 1 shows the calculation formulas for the time-domain features used in this step, and Table 2 shows the calculation formulas for the frequency-domain features used in this invention. The sensor signal data is X = {x} i Let i = 1, 2, 3, ..., I. The square of the amplitude spectrum of signal x(t) is |X(ω)|. 2 The resulting sequence is called the power spectrum S(ω) of the signal.

[0069] Table 1

[0070]

[0071]

[0072] Table 2

[0073]

[0074] Step 1.3 Select sensor signal characteristics that are sensitive to but not redundant to changes in the average width VB of the wear band;

[0075] The features extracted in step 1.2 do not all clearly reflect the changing trend of tool wear values, and excessive information can lead to chaotic clustering results. Therefore, it is necessary to select features that are sensitive to changes in the average wear band width VB but are not redundant. This process is called feature selection. This invention combines the Pearson correlation coefficient method, the mutual information coefficient method, and the fuzzy C-means clustering method to select features that are sensitive to the evolution of the average wear band width VB.

[0076] (1) The formula for calculating the Pearson correlation coefficient:

[0077]

[0078] In the formula: ρ mn The Pearson correlation coefficient is used to compare characteristics M = {m} i The average width VB of the wear band is N = {n}, where i = 1, 2, 3, ..., I}. i (i = 1, 2, 3, ..., I).

[0079] Calculate the Pearson correlation coefficient between each feature extracted in step 1.2 and the average width VB of the wear band. Set the correlation coefficient threshold to 0.9 and select features that are significantly correlated with the average width VB of the wear band.

[0080] (2) The formula for calculating the mutual information coefficient is:

[0081]

[0082] In the formula: I(N, M) is the mutual information coefficient, P(m) is the mutual information coefficient. i n j ) is m i and n j The joint distribution probability.

[0083] Calculate the mutual information coefficient between each feature extracted in step 1.2 and the average width VB of the wear band. Set the threshold for the mutual information coefficient to 0.85 and select features that are significantly correlated with the average width VB of the wear band.

[0084] (3) Fuzzy C-means clustering

[0085] Cluster analysis divides data points into different clusters according to certain criteria, ensuring maximum similarity among data points within the same cluster and minimum similarity among data points in different clusters. The fuzzy C-means clustering algorithm differs from common clustering algorithms such as K-means clustering in that K-means rigidly assigns data points to a specific cluster, while fuzzy C-means incorporates the concept of fuzziness. It no longer absolutely assigns data points to a particular cluster, but rather considers them to belong to all clusters simultaneously, using a membership degree between [0, 1] to represent the degree to which a data point belongs to a particular cluster.

[0086] Using the fuzzy C-means clustering algorithm, the average wear band width VB was combined with 92 feature vectors to form 93 data points, which were then divided into two clusters. Features in the same cluster as the average wear band width VB were considered to have a strong correlation with tool wear, and their membership degree was used to represent the magnitude of this correlation. Features in the other cluster were considered to be unrelated to tool wear evolution. A membership degree threshold of 0.8 was set to select features with a significantly high correlation to the average wear band width VB.

[0087] Finally, the intersection of the features selected by the Pearson correlation coefficient method, the mutual information coefficient method, and the fuzzy C-means clustering method is taken as the final selected features.

[0088] Step 1.4 Construction and normalization of the feature matrix;

[0089] Since different features have different dimensions and orders of magnitude, and the order of magnitude difference between features is large, directly using the features selected in step 1.3 to construct a feature matrix and inputting the feature matrix into the clustering analysis model will weaken the influence of features with lower orders of magnitude. In order to extract the trend of the selected features and discard the influence of their dimensions and orders of magnitude on the clustering results, it is necessary to normalize the selected features and compress the feature values ​​to the range of [0, 1].

[0090] In this step, a feature matrix of size T = P × Q is constructed as the input to the subsequent evolutionary clustering model, where P is the number of average wear band width VB values ​​in the experimental group, and Q is the number of features selected in step 1.3.

[0091] Step 1.5 Evolutionary cluster analysis to obtain a series of time series tool wear evolution cluster distribution maps;

[0092] In this step, cluster analysis is first performed sequentially on the first few feature matrices. For each cluster analysis, agglomerative hierarchical clustering is used, and the Calinski-Harabasz index, Davies-Bouldin index, and Average Silhouette Method silhouette coefficient are selected to determine the optimal number of clusters for each clustering. Then, principal component analysis is used to reduce the dimensionality of the feature matrices obtained in step 1.4 to visualize the evolutionary cluster analysis results. Finally, a series of time-series cluster distribution maps of tool wear evolution are obtained, some examples of which are shown below. Figure 7 As shown.

[0093] Step 2: Perform a consistency analysis on the tool wear evolution cluster distribution map obtained in Step 1 to obtain a standard cluster distribution map.

[0094] Step 2.1 Count the number of deviations of each data point from the standard cluster;

[0095] In the cluster distribution diagram, all data points are distributed in different clusters. As the number of cuts increases, the number of data points representing the feature matrix of the monitoring signal also increases. New data points are added to existing clusters or form new clusters; existing data points will belong to existing clusters during the evolutionary cluster analysis process.

[0096] In this step, we first count the total number of occurrences of all data points in the tool wear evolution clustering distribution map over the time series. And to count the mode of each data point in the time series, which it belongs to each cluster. For any data point, the cluster with the highest membership frequency is taken as the standard cluster for that data point, thus obtaining the standard clusters corresponding to each data point, whose distribution is as follows: Figure 8 As shown; based on Figure 8 Furthermore, the number of standard cluster categories corresponding to different cutting times can be obtained;

[0097] Then, for each data point, count the number of deviations that data point makes from its corresponding standard cluster. The formula for calculating the number of deviations is:

[0098]

[0099] Table 3 shows examples of the total number of occurrences of data points corresponding to the number of cuts, the mode of each data point belonging to each cluster in the time series, and the number of deviations of data points from the standard cluster.

[0100] Table 3

[0101]

[0102] Step 2.2 Calculate the deviation coefficient EC;

[0103] The deviation coefficient EC is used to describe the consistency of the tool wear evolution cluster distribution map obtained in step 1; it represents the number of deviations from all data points. The sum of the total number of occurrences of all data points The ratio is called the deviation coefficient EC, and the formula for calculating the deviation coefficient EC is:

[0104]

[0105] Step 2.3 Obtain a series of standard cluster distribution maps for evolutionary cluster distribution maps;

[0106] A deviation coefficient EC greater than 85% is considered to indicate good consistency in the evolutionary clustering distribution map. In this case, the standard clusters of each data point obtained in step 2.1 (as shown in Table 3) can be placed in the same graph to construct a standard clustering distribution map. This constructed standard clustering distribution map is then used as... Figure 7 Simplification of the evolutionary clustering distribution map in, such as Figure 8 As shown.

[0107] If the deviation coefficient EC of the evolutionary cluster distribution map is less than or equal to 85%, it is necessary to return to step 1.3 to re-screen features and execute the steps after step 1.3 until the deviation coefficient EC of the evolutionary cluster distribution map is greater than 85%, so as to ensure that more obvious stage features can be obtained in the future, so that stage analysis can be performed.

[0108] Step 3: Based on the stage characteristics of the standard cluster distribution map obtained in Step 2, perform stage analysis on it to divide it into multiple different continuous intervals that reflect different stages of the tool wear evolution process, and calculate the cluster density of each continuous interval.

[0109] The standard cluster distribution diagram exhibits distinct phased characteristics: at the beginning of the cutting process, there are many clusters but few data points within each cluster; in the middle of the cutting experiment, there are few clusters but many data points within each cluster; and at the end of the cutting experiment, there are many clusters but few data points within each cluster. Therefore, based on these phased characteristics, the standard cluster distribution diagram can be divided into several distinct continuous intervals.

[0110] Then, in order to evaluate the dispersion of clusters within different continuous intervals in the standard cluster distribution map, the cluster density ρ of each continuous interval is calculated. s～e To describe the continuous interval (from data point s to data point e), and the corresponding cluster categories o s to o e Within the range, the density of different data points belonging to different clusters represents the intensity of the evolutionary process.

[0111] Cluster density ρ s～e The calculation formula is:

[0112]

[0113] In summary, by calculating the cluster density ρ of different continuous intervals... s～e This allows us to distinguish continuous intervals with different degrees of evolutionary intensity by using numerical values. For example, in this embodiment, the data point ranges for different stages are 1-5, 6-14, 15-28, and 29-40, with corresponding cluster densities of 1, 0.111, 0.357, and 0.75, respectively.

[0114] Step 4: Perform correlation analysis on the multiple different continuous intervals obtained in Step 3 to determine whether they can reflect the evolution process of tool wear.

[0115] Step 4.1 Calculate the average characteristic rate of the normalized characteristic matrix obtained from the sensor signal at different stages. and average wear rate

[0116] In this step, to verify whether the multiple different continuous intervals obtained from step 3 can reflect the evolution of tool wear, firstly, the different continuous intervals obtained in step 3 (from data point s to data point e, eigenvalue f) are calculated. e to f s The average characteristic rate of the normalized characteristic matrix obtained from the sensor signal in ) In this embodiment, the values ​​are 0.0300, 0.0178, 0.0210, and 0.0254, respectively; and the average wear rate in different stages obtained in step 3 is calculated. In this embodiment, the values ​​are 0.0121, 0.0046, 0.0049, and 0.0097 (mm / cut), respectively.

[0117] Average wear rate The calculation formula is:

[0118]

[0119] In the formula: VB e VB s-1 These represent the tool wear values ​​corresponding to data points s-1 to e within the continuous interval;

[0120] Average characteristic rate The calculation formula is:

[0121]

[0122] In the formula: f e f s-1 These represent the characteristic values ​​corresponding to data points s-1 to e within the continuous interval.

[0123] Then, plot the average characteristic rate. With cluster density ρ s～e The comparison diagram is shown in this embodiment. Figure 9 As shown, the average wear rate is plotted. With cluster density ρ s～e The comparison diagram is shown in this embodiment. Figure 10 As shown. Visual. Figure 9 and Figure 10 This can corroborate that the evolution trend of cluster density is the same as that of average characteristic rate and average wear rate, and can also serve as a reference for the selection of subsequent sub-functions.

[0124] Step 4.2 Calculate the correlation between cluster density and average characteristic rate, and between cluster density and average wear rate;

[0125] To evaluate the cluster density ρ of different continuous intervals s～e and average characteristic rate The correlation between them, and the cluster density ρ s～e and average wear rate The correlation between them was calculated, and the Pearson correlation coefficient and mutual information coefficient were determined. A good correlation was considered when the Pearson correlation coefficient and mutual information coefficient for all continuous intervals were greater than 0.85, indicating that the different continuous intervals divided in step 3 matched the changing trends of the evolution data (sensor signals and the average width VB value of the wear band) during the machining process. Subsequently, the cluster density of different continuous intervals could be used for tool wear modeling. In this embodiment, their Pearson correlation coefficients were 0.9947 and 0.9604, and their mutual information coefficients were 1 and 1, respectively, both greater than 0.85, indicating a good correlation.

[0126] If either the Pearson correlation coefficient or the mutual information coefficient is less than or equal to 0.85, the correlation is considered poor. It is necessary to return to step 1.3 to re-select features and execute the steps after step 1.3 until both the Pearson correlation coefficient and the mutual information coefficient are greater than 0.85. This ensures a high correlation between cluster density and average feature rate, and between cluster density and average wear rate, serving as evidence that multiple different continuous intervals can reflect the evolution of tool wear.

[0127] Step 5: Construct a tool wear model.

[0128] Step 5.1: Construct basic constraints;

[0129] In this step, based on the physical properties of tool wear, the basic constraints that the tool wear model should meet are first given:

[0130]

[0131] In the formula, VB is the average width of the wear band; ΔVB is the differential form of VB; t is the cutting time; Δt is the differential form of t; t0 is the approximate value of t; w(t) is the tool wear model, which satisfies VB=w(t); w′(t) is the first derivative of w(t); w″(t) is the second derivative of w(t).

[0132] Step 5.2: Construct the tool wear model:

[0133] Under the premise of satisfying the basic constraints established in step 5.1, the tool wear model must accurately fit the tool wear curve. This requires the tool wear model to be composed of different sub-function equations to adapt to the different stages of tool wear evolution given in the above equation. The tool wear model is composed of sub-equations:

[0134] w(t) = w 1 (t)+w 2 (t)+......+w k (t)

[0135] In addition to satisfying the basic constraints established in step 5.1, each sub-function equation should also satisfy the following requirements:

[0136] a) Different sub-equations control different evolutionary stages;

[0137] b) The slopes of different sub-equations in other evolutionary stages are less than the slopes in their respective evolutionary stages.

[0138] Based on the trend charts of cluster density and average characteristic rate for different continuous intervals obtained in step 4 Figure 9 And trend graphs of cluster density and average tool wear rate at different stages. Figure 10 Based on the slope variation characteristics of various basic functions such as power functions, exponential functions, logarithmic functions, polynomial functions, and constant functions, specific functions are selected to control the evolution stages of different continuous intervals. For example, the exponential function has the characteristic of a rapid slope change in the early stage and a slope that is basically zero in the remaining stages, which can control the evolution stage where the early evolution is more intense. The selection of different sub-equations is based on requirements a) and b). Ultimately, this embodiment selects exponential functions, power functions, and linear functions to construct the tool wear model. Among them, the exponential function is used to control the process of the slope of the curve decreasing from large to small in the early stage, and then the slope converges to zero, without affecting the slope of the subsequent curve; the power function is used to control the process of the slope gradually increasing from stable in the subsequent stage, with a relatively small impact on the slope of the early curve; the linear function is used to adjust the initial value of the wear curve and control the overall position of the curve.

[0139]

[0140] The final tool wear model is as follows:

[0141] w(t) = w 1 (t)+w 2 (t)+w 3 (t)=-ae -bt +cJ t +et+f

[0142] In the formula, a, b, c, d, e, and f are the fitting coefficients of the tool wear model, and satisfy a, b, c, e > 0, d > 1, and f ≥ ace. When the tool is not worn during cutting, f = ace is satisfied.

[0143] Verification of the effect of this embodiment:

[0144] The tool wear model constructed in step 5 of this embodiment is used to fit the tool wear curve, and the result is as follows: Figure 11 As shown. From Figure 11 It can be seen that the tool wear model can fit the data very well. Figure 3The evolution process of the GH4169 tool wear curve obtained from the machining experiment is shown, and its sub-function equation w 1 w 2 and w 3 The curve slope and position were effectively controlled in the early, late, and overall stages. The fitting accuracy of w(t) was as high as 97.843%, and the coefficient of determination R² was as high as 99.566%, indicating that the tool wear model constructed in this embodiment of the invention has good applicability to the tool wear curve of GH4169.

[0145] The following experiments compare the fitting capabilities of three existing tool wear models with the tool wear model proposed in this invention. The three existing tool wear models w2(t), w3(t), and w4(t) are as follows:

[0146]

[0147]

[0148]

[0149] In the formula a 2～4 b 2～4 and c 3～4 These are the model's fit coefficients.

[0150] The fitting results of the tool wear models w2(t), w3(t), and w4(t) are as follows: Figure 12 , 13 As shown in Figures 1 and 14, the fitting accuracies were 91.447%, 94.648%, and 96.094%, respectively; the coefficients of determination (R²) were 94.656%, 97.483%, and 98.592%, respectively. These figures are all lower than the indices of the tool wear model constructed in the embodiments of this invention, indicating that this invention can obtain a tool wear model with better fitting ability than existing tool wear models.

[0151] In addition to the tool wear modeling method based on evolutionary clustering analysis described above, this invention also provides a non-volatile computer-readable storage medium storing a computer program thereon, which, when run, is used to execute the tool wear modeling method based on evolutionary clustering analysis of this invention.

[0152] Meanwhile, the present invention also provides an electronic device, including a processor and a storage medium, wherein a computer program is stored on the storage medium, and the computer program is executed by the processor to perform the tool wear modeling method based on evolutionary clustering analysis of the present invention.

Claims

1. A tool wear modeling method based on evolutionary clustering analysis, including Step 1: Obtain a series of time-series cluster distribution maps of tool wear evolution; Step 1.1 Collect raw evolutionary data; Step 1.2: Extract features from the sensor signals; Step 1.3 Select sensor signal characteristics that are sensitive to but not redundant to changes in the average width VB of the wear band; Step 1.4 Construct a feature matrix using the sensor signal features selected in Step 1.3 and normalize it; Step 1.5 involves performing evolutionary cluster analysis on the feature matrix obtained in Step 1.4 to obtain a series of time-series tool wear evolution cluster distribution maps; Its characteristic is that it further includes the following steps: Step 2: For each data point in the tool wear evolution clustering distribution map, obtain its standard cluster and the number of deviations from its corresponding standard cluster, and then calculate the deviation coefficient EC of the evolution clustering distribution map. If EC is greater than the set value, the consistency is considered good. At this time, the standard clustering distribution map is constructed from the standard clusters of each data point, and proceed to step 3; otherwise, the consistency is considered poor, and return to step 1.

3. Step 3: Based on the stage-specific characteristics of the standard cluster distribution map, divide it into multiple distinct continuous intervals to reflect different stages of the tool wear evolution process, and calculate the cluster density ρ of each continuous interval. s～e ; s and e are the starting and ending points of the data in the continuous interval s to e, respectively; Step 4: For the multiple different continuous intervals, calculate the average feature rate of the normalized feature matrix obtained from the sensor signal. and average wear rate Perform a correlation analysis with the cluster density. If the correlation meets the requirements, proceed to step 5; otherwise, return to step 1.

3. Step 5: Construct the tool wear model: Step 5.1 Construct basic constraints: In the formula, VB is the average width of the wear band; ΔVB is the differential form of VB; t is the cutting time; Δt is the differential form of t; t0 is the approximation value of t; w(t) is the tool wear model, which satisfies VB=w(t); w′(t) is the first derivative of w(t); w″(t) is the second derivative of w(t); Step 5.2: Construct the tool wear model w(t): w(t)=w 1 (t)+w 2 (t)+......+w k (t) In the formula, w 1 (t), w 2 (t), ..., w k (t) represent the sub-equations corresponding to different continuous intervals, and each sub-equation is based on the cluster density and average characteristic rate of the different continuous intervals. The trend charts of cluster density and average tool wear rate in different continuous intervals are determined by combining the slope variation characteristics of various basic functions. Each sub-equation satisfies the basic constraints constructed in step 5.1, and the requirements are: a) different sub-equations control different evolution stages; b) the slope of different sub-equations in other evolution stages is less than the slope in their respective evolution stages.

2. The tool wear modeling method based on evolutionary clustering analysis according to claim 1, characterized in that: The specific method for obtaining the deviation coefficient EC in step 2 is as follows: Step 2.1 Count the number of deviations of each data point from the standard cluster; Step 2.1.1 Calculate the total number of occurrences of all data points in the cluster distribution map of tool wear evolution over a statistical time series. The mode of each data point in the time series, belonging to its respective cluster. For any data point, the cluster with the most memberships is taken as the standard cluster for that data point; Step 2.1.2 For each data point, count the number of deviations that data point makes from its corresponding standard cluster. Step 2.2 Calculate the deviation coefficient EC: In the formula, This is the sum of the number of deviations for all data points; This represents the total number of occurrences of all data points.

3. The tool wear modeling method based on evolutionary clustering analysis according to claim 1, characterized in that: The cluster density of each consecutive interval in step 3 is calculated according to the following formula: In the formula, o e and o s These represent the cluster categories to which the starting point and ending point of the data in the continuous interval s to e belong, respectively.

4. The tool wear modeling method based on evolutionary clustering analysis according to claim 1, characterized in that: The average wear rate of each continuous interval in step 4 Calculate according to the following formula: The average characteristic rate is calculated according to the following formula:

5. The tool wear modeling method based on evolutionary clustering analysis according to claim 1, characterized in that: In step 4, the average feature rate of the normalized feature matrix obtained from the sensor signal is... and average wear rate The method for performing correlation analysis with the cluster density is as follows: For each continuous interval, calculate the average characteristic rate. With cluster density ρ s～e The Pearson correlation coefficient and mutual information coefficient, and the calculation of the average wear rate. With cluster density ρ s～e Pearson correlation coefficient and mutual information coefficient; When the Pearson correlation coefficient and mutual information coefficient corresponding to all continuous intervals are greater than the set threshold, it indicates that the correlation meets the requirements. At this time, it is considered that the different continuous intervals divided in step 3 are consistent with the changing trend of the evolution data in the processing process.

6. A non-volatile computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is run, it is used to execute the tool wear modeling method based on evolutionary clustering analysis as described in any one of claims 1-5.

7. An electronic device, comprising a processor and a storage medium, wherein a computer program is stored on the storage medium, characterized in that: When the computer program is run by the processor, it is used to execute the tool wear modeling method based on evolutionary clustering analysis as described in any one of claims 1-5.

Images