A real-time perception method for abnormal cutting conditions of ceramic matrix composites based on PCD tools

Abstract

The present invention provides a real-time detection method for abnormal cutting conditions in ceramic-matrix composites using PCD tools. The method includes obtaining a preferred combination of cutting parameters for ceramic-matrix composites, establishing a tool loss prediction model, calculating an estimated tool loss value, and identifying abnormal cutting conditions in ceramic-matrix composites. This method provides real-time feedback on the status of the PCD tool's cutting conditions for ceramic-matrix composites, predicting changing cutting conditions, improving the timeliness of response to condition degradation, and ultimately enhancing the machining quality of ceramic-matrix composites.

Description

Technical Field

[0001] The present invention relates to the field of tool wear prediction, and in particular to a real-time perception method for abnormal cutting conditions of ceramic-based composite materials based on PCD tools. Background Art

[0002] PCD tools are commonly used in the precision machining of ceramic-matrix composite (CM) workpieces. PCD tool breakage and micro-chipping are direct causes of surface degradation and reduced machining accuracy. Effectively identifying tool breakage and chipping and making timely preventative process decisions are key to ensuring workpiece quality and accuracy.

[0003] In actual engineering, experience-based tool wear and working condition identification is the most commonly used method. Process adjustments are made by utilizing the sensory changes of workpiece vibration, cutting sound and the macroscopic quality of the cutting surface. However, this method lacks objectivity and has lags, which are specifically reflected in the following aspects: ① Experience-based process adjustments lack unified principles and standards, and the deviation of working condition identification is large, which easily leads to the deterioration of processing quality due to untimely process adjustments; ② The macroscopic quality of the workpiece surface is an important criterion for the working condition. Once the surface quality deteriorates, the tool breakage is often already quite serious, and the processing accuracy has been seriously affected.

[0004] Therefore, it is of great significance to develop a real-time perception method for abnormal cutting conditions of ceramic matrix composites based on PCD tools. Summary of the Invention

[0005] The purpose of the present invention is to provide a real-time perception method for abnormal cutting conditions of ceramic-based composite materials based on PCD tools, so as to solve the problems existing in the prior art.

[0006] The technical solution adopted to achieve the purpose of the present invention is as follows: a real-time perception method for abnormal cutting conditions of ceramic matrix composite materials based on PCD tools, comprising the following steps:

[0007] 1) Carry out orthogonal cutting tests on ceramic matrix composites based on PCD tools to obtain the optimal cutting parameter combination {S, V, a w ,a p}. Among them, S represents the spindle speed, V represents the feed speed, a w Indicates the cutting width, a p Indicates the cutting depth.

[0008] 2) Based on the cutting parameter combination obtained in step 1), a single-factor cutting test on cutting length was carried out to establish a tool loss value prediction model based on a shallow neural network:

[0009] u=Φ(D′)

[0010] Where D′ is the time-frequency domain feature data matrix of the cutting condition after dimensionality reduction, and u is the column vector of tool loss.

[0011] 3) Real-time acquisition of working condition signals during the cutting process, and the use of the established tool loss value prediction model to calculate the tool loss estimate u e .

[0012] 4) Determine the estimated value of tool loss u e and the preset loss threshold u b The relationship between the size of the PCD tool and the abnormal cutting conditions of the ceramic matrix composite material can be identified. e Greater than the preset loss threshold u b When the cutting speed is 0.05, it is determined as an abnormal working condition of ceramic matrix composite material cutting; otherwise, it is determined as a normal working condition.

[0013] Furthermore, step 1) specifically includes the following sub-steps:

[0014] 1.1) Orthogonal cutting tests of ceramic matrix composites using PCD cutting tools were conducted. The radial vibration data of the cutting tool during each cutting test were collected at a certain sampling frequency to obtain a time domain working condition dataset.

[0015] 1.2) After each set of cutting tests, the thickness of the PCD material layer on the flank surface of the tool's main cutting edge and secondary cutting edge was measured as the tool's main and secondary cutting edge loss values.

[0016] 1.3) Carry out range analysis based on the loss values ​​of the primary and secondary cutting edges respectively to obtain the cutting parameter combination based on the principle of minimum loss values ​​of the primary and secondary cutting edges, including spindle speed, feed rate, cutting width and cutting depth.

[0017] Furthermore, in step 2), a single factor cutting test with equidistantly increasing cutting length is carried out.

[0018] Furthermore, step 2) specifically includes the following sub-steps:

[0019] 2.1) Based on the obtained optimal cutting parameter combination, single-factor cutting tests with equidistant increasing cutting lengths were carried out. The radial vibration data of the tool during each cutting test was collected at a certain sampling frequency to obtain a time domain working condition data set for different cutting lengths.

[0020] 2.2) After each set of cutting tests, the thickness of the material layer damaged on the flank face of the tool’s main cutting edge was measured as the tool loss value for the current cutting length. A new tool was then replaced for the next set of tests until the tool loss value after a set of cutting tests exceeded 300 μm. The data was then summarized to obtain a tool loss column vector.

[0021] 2.3) Time domain analysis is performed on the time domain working condition data set to obtain the time domain feature data set of different cutting lengths. The data attributes of each element of the data set include average value, rectified average value, maximum value, minimum value, peak value, valley value, peak-to-peak value, standard deviation and root mean square.

[0022] 2.4) A discrete Fourier transform is performed on the time domain working condition data set to obtain a frequency domain feature data set for different cutting lengths. The data attributes of each element in the data set include maximum amplitude, maximum amplitude frequency, center of gravity frequency, average amplitude, total energy in the frequency band, and the proportion of low-frequency energy from 40 to 200 Hz.

[0023] 2.5) Combine the time domain feature data set and the frequency domain feature data set of the same cutting length to obtain the time-frequency domain feature data row vector of the cutting condition with a certain cutting length, and combine the time-frequency domain feature data row vectors of the condition in the increasing order of the cutting length to obtain the time-frequency domain feature data matrix of the cutting condition.

[0024] 2.6) By analyzing the correlation between each column vector in the cutting condition time-frequency domain characteristic data matrix and the tool loss column vector, the reduced-dimensional cutting condition time-frequency domain characteristic data matrix is ​​obtained.

[0025] 2.7) Using the reduced-dimensional cutting condition time-frequency domain feature data matrix as input and the tool loss column vector as output, a tool loss value prediction model based on a shallow neural network is established.

[0026] Further, step 2.6) specifically includes the following sub-steps:

[0027] 2.6.1) Calculate the Spearman correlation coefficient between each column vector of the cutting condition time-frequency domain characteristic data matrix and the tool loss column vector in sequence to obtain the correlation coefficient row vector.

[0028] 2.6.2) Construct a statistic that follows a standard normal distribution and, using the obtained Spearman correlation coefficient row vector, calculate each test value in turn to obtain a test value row vector.

[0029] 2.6.3) Using the obtained test value row vector, calculate each P value in turn to obtain a P value row vector.

[0030] 2.6.4) Determine the significance level of the correlation between the corresponding column vectors of the cutting condition time-frequency domain characteristic data matrix and the tool wear column vector based on the P-value row vectors, and mark the redundant column vectors.

[0031] 2.6.5) Remove redundant column vectors from the cutting condition time-frequency domain feature data matrix to obtain a reduced-dimensional cutting condition time-frequency domain feature data matrix.

[0032] Furthermore, step 3) specifically includes the following sub-steps:

[0033] 3.1) The radial vibration data of the tool during the cutting process is collected in real time at a certain sampling frequency, and the data is summarized at a given time interval to obtain a segmented time domain working condition data set within a certain sampling interval.

[0034] 3.2) Perform time domain analysis on the segmented time domain operating condition data set to obtain a segmented time domain feature data set. The data attributes of each element in the data set include average value, rectified average value, maximum value, minimum value, peak value, valley value, peak-to-peak value, standard deviation, and root mean square.

[0035] 3.3) Perform discrete Fourier transform on the segmented time-domain operating condition dataset to obtain a segmented frequency-domain feature dataset. The data attributes of each element in the dataset include maximum amplitude, maximum amplitude frequency, center of gravity frequency, average amplitude, total energy in the frequency band, and the proportion of low-frequency energy from 40 to 200 Hz.

[0036] 3.4) Combine the segmented time domain feature data set and the frequency domain feature data set to obtain the segmented time and frequency domain feature data row vector of the cutting condition.

[0037] 3.5) Performing dimensionality reduction processing on the row vectors of the segmented time-frequency domain characteristic data of the cutting condition, removing the elements in the row vectors that have the same column numbers as the redundant column vectors removed by the dimensionality reduction process of the time-frequency domain characteristic data matrix of the cutting condition in step 2.6).

[0038] 3.6) Taking the time-frequency domain feature row vector of the cutting condition after dimensionality reduction as input, the tool loss estimation value is calculated using the established tool loss value prediction model.

[0039] The technical benefits of this invention are undeniable: it provides real-time feedback on the cutting conditions of ceramic-matrix composites using PCD tools, predicting changing cutting conditions, improving the timeliness of response to degradation, and enhancing the machining quality of ceramic-matrix composites. By analyzing and processing data from vibration signals from cutting ceramic-matrix composites using PCD tools, tool wear can be predicted in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of a real-time perception method for abnormal cutting conditions of ceramic matrix composites based on PCD tools;

[0041] Figure 2 Schematic diagram of dimensionality reduction processing of the time-frequency domain characteristic data matrix of cutting conditions based on Spearman correlation analysis. DETAILED DESCRIPTION

[0042] The present invention will be further described below with reference to the following examples, but it should not be understood that the scope of the present invention is limited to the following examples. Without departing from the above technical ideas of the present invention, various substitutions and modifications can be made according to common technical knowledge and customary means in the art, and all should be included in the scope of protection of the present invention.

[0043] Example 1:

[0044] This embodiment provides a method for real-time sensing of abnormal working conditions in cutting ceramic matrix composite materials based on PCD cutting tools, comprising the following steps:

[0045] 1) Carry out orthogonal cutting tests on ceramic matrix composites based on PCD tools to obtain the optimal cutting parameter combination {S, V, a w ,a p}. Among them, S represents the spindle speed, V represents the feed speed, a w Indicates the cutting width, a p Indicates the cutting depth.

[0046] 2) Based on the cutting parameter combination obtained in step 1), a single-factor cutting test on cutting length was carried out to establish a tool loss value prediction model based on a shallow neural network:

[0047] u=Φ(D′)

[0048] Where D′ is the time-frequency domain feature data matrix of the cutting condition after dimensionality reduction, and u is the column vector of tool loss.

[0049] 3) Real-time acquisition of working condition signals during the cutting process, and the use of the established tool loss value prediction model to calculate the tool loss estimate u e .

[0050] 4) Determine the estimated value of tool loss u e and the preset loss threshold u b The relationship between the size of the PCD tool and the abnormal cutting conditions of the ceramic matrix composite material can be identified. e Greater than the preset loss threshold u b When the cutting speed is 0.05, it is determined as an abnormal working condition of ceramic matrix composite material cutting; otherwise, it is determined as a normal working condition.

[0051] Example 2:

[0052] The main contents of this embodiment are the same as those of embodiment 1, wherein step 1) specifically includes the following sub-steps:

[0053] 1.1) Orthogonal cutting tests of ceramic matrix composites using PCD cutting tools were conducted. The radial vibration data of the cutting tool during each cutting test were collected at a certain sampling frequency to obtain a time domain working condition dataset.

[0054] 1.2) After each set of cutting tests, the thickness of the PCD material layer on the flank surface of the tool's main cutting edge and secondary cutting edge was measured as the tool's main and secondary cutting edge loss values.

[0055] 1.3) Carry out range analysis based on the loss values ​​of the primary and secondary cutting edges respectively to obtain the cutting parameter combination based on the principle of minimum loss values ​​of the primary and secondary cutting edges, including spindle speed, feed rate, cutting width and cutting depth.

[0056] Example 3:

[0057] The main contents of this embodiment are the same as those of embodiment 1 or 2, wherein in step 2), a single-factor cutting test with equidistantly increasing cutting length is carried out. Step 2) specifically includes the following sub-steps:

[0058] 2.1) Based on the obtained optimal cutting parameter combination, single-factor cutting tests with equidistant increasing cutting lengths were carried out. The radial vibration data of the tool during each cutting test was collected at a certain sampling frequency to obtain a time domain working condition data set for different cutting lengths.

[0059] 2.2) After each set of cutting tests, the thickness of the material layer damaged on the flank face of the tool’s main cutting edge was measured as the tool loss value for the current cutting length. A new tool was then replaced for the next set of tests until the tool loss value after a set of cutting tests exceeded 300 μm. The data was then summarized to obtain a tool loss column vector.

[0060] 2.3) Time domain analysis is performed on the time domain working condition data set to obtain the time domain feature data set of different cutting lengths. The data attributes of each element of the data set include average value, rectified average value, maximum value, minimum value, peak value, valley value, peak-to-peak value, standard deviation and root mean square.

[0061] 2.4) A discrete Fourier transform is performed on the time domain working condition data set to obtain a frequency domain feature data set for different cutting lengths. The data attributes of each element in the data set include maximum amplitude, maximum amplitude frequency, center of gravity frequency, average amplitude, total energy in the frequency band, and the proportion of low-frequency energy from 40 to 200 Hz.

[0062] 2.5) Combine the time domain feature data set and the frequency domain feature data set of the same cutting length to obtain the time-frequency domain feature data row vector of the cutting condition with a certain cutting length, and combine the time-frequency domain feature data row vectors of the condition in the increasing order of the cutting length to obtain the time-frequency domain feature data matrix of the cutting condition.

[0063] 2.6) By analyzing the correlation between each column vector in the cutting condition time-frequency domain characteristic data matrix and the tool wear column vector, a dimensionally reduced cutting condition time-frequency domain characteristic data matrix is ​​obtained. Step 2.6) specifically includes the following sub-steps:

[0064] 2.6.1) Calculate the Spearman correlation coefficient between each column vector of the cutting condition time-frequency domain characteristic data matrix and the tool loss column vector in sequence to obtain the correlation coefficient row vector.

[0065] 2.6.2) Construct a statistic that follows a standard normal distribution and, using the obtained Spearman correlation coefficient row vector, calculate each test value in turn to obtain a test value row vector.

[0066] 2.6.3) Using the obtained test value row vector, calculate each P value in turn to obtain a P value row vector.

[0067] 2.6.4) Determine the significance level of the correlation between the corresponding column vectors of the cutting condition time-frequency domain characteristic data matrix and the tool wear column vector based on the P-value row vectors, and mark the redundant column vectors.

[0068] 2.6.5) Remove redundant column vectors from the cutting condition time-frequency domain feature data matrix to obtain a reduced-dimensional cutting condition time-frequency domain feature data matrix.

[0069] 2.7) Using the reduced-dimensional cutting condition time-frequency domain feature data matrix as input and the tool loss column vector as output, a tool loss value prediction model based on a shallow neural network is established.

[0070] Example 4:

[0071] The main contents of this embodiment are the same as those of embodiment 3, wherein step 3) specifically includes the following sub-steps:

[0072] 3.1) The radial vibration data of the tool during the cutting process is collected in real time at a certain sampling frequency, and the data is summarized at a given time interval to obtain a segmented time domain working condition data set within a certain sampling interval.

[0073] 3.2) Perform time domain analysis on the segmented time domain operating condition data set to obtain a segmented time domain feature data set. The data attributes of each element in the data set include average value, rectified average value, maximum value, minimum value, peak value, valley value, peak-to-peak value, standard deviation, and root mean square.

[0074] 3.3) Perform discrete Fourier transform on the segmented time-domain operating condition dataset to obtain a segmented frequency-domain feature dataset. The data attributes of each element in the dataset include maximum amplitude, maximum amplitude frequency, center of gravity frequency, average amplitude, total energy in the frequency band, and the proportion of low-frequency energy from 40 to 200 Hz.

[0075] 3.4) Combine the segmented time domain feature data set and the frequency domain feature data set to obtain the segmented time and frequency domain feature data row vector of the cutting condition.

[0076] 3.5) Performing dimensionality reduction processing on the row vectors of the segmented time-frequency domain characteristic data of the cutting condition, removing the elements in the row vectors that have the same column numbers as the redundant column vectors removed by the dimensionality reduction process of the time-frequency domain characteristic data matrix of the cutting condition in step 2.6).

[0077] 3.6) Taking the time-frequency domain feature row vector of the cutting condition after dimensionality reduction as input, the tool loss estimation value is calculated using the established tool loss value prediction model.

[0078] Example 5:

[0079] The main contents of this embodiment are the same as those of embodiment 1, wherein, see Figure 1 and Figure 2 , this embodiment includes the following steps:

[0080] 1) Conduct orthogonal cutting tests on ceramic matrix composites using PCD tools to obtain the optimal cutting parameter combination for ceramic matrix composites. Step 1) specifically includes the following sub-steps:

[0081] 1.1) According to the preset orthogonal experimental factor level table, n groups of ceramic matrix composite cutting tests using PCD tools were planned. The radial vibration data of the tool during the first to n groups of cutting tests were collected in sequence at a certain sampling frequency f to obtain a time domain working condition data set:

[0082]

[0083] in, It represents the amplitude of the tool radial vibration acceleration at the kth sampling moment in the i-th group of cutting test process.

[0084] 1.2) After each set of cutting tests, the thickness of the PCD material layer at the flank of the tool's main cutting edge and secondary cutting edge was measured as the tool's main and secondary cutting edge loss values:

[0085] {u1,u2,u3,...,u i ,...,u n}

[0086] {v1,v2,v3,...,v i ,...,v n}

[0087] Among them, u i represents the damaged thickness of the PCD material layer on the flank face of the main cutting edge after the i-th group of cutting tests, v i It represents the damaged thickness of the PCD material layer on the flank face of the secondary cutting edge after the i-th group of cutting tests.

[0088] 1.3) Carry out the range analysis based on the loss value of the main and secondary cutting edges respectively, and obtain the cutting parameter combination {S, V, a w ,a p}, where S represents the spindle speed, V represents the feed speed, a w Indicates the cutting width, a p Indicates the cutting depth.

[0089] 2) Conduct a single-factor cutting test on cutting length and establish a PCD tool loss prediction model, which includes the following steps:

[0090] 2.1) Based on the obtained preferred cutting parameter combination {S, V, a w ,a p}, carry out a single factor cutting test with equidistant increasing cutting length, take the cutting length of the first group of tests as L0, the increasing value of the cutting length of each group of tests as ΔL, and the cutting length of the i-th group of tests as L i , L i = L0 + (i-1) * ΔL. The radial vibration data of the tool during each cutting test is collected at a certain sampling frequency f to obtain the time domain working condition data set of different cutting lengths:

[0091]

[0092] in, It represents the amplitude of the tool radial vibration acceleration at the kth sampling moment in the i-th group of cutting test process.

[0093] 2.2) After each set of cutting tests, the thickness of the material layer damaged on the flank face of the tool’s main cutting edge is measured as the tool loss value for the current cutting length. A new tool is then replaced for the next set of tests. This process continues until the tool loss value after a set of cutting tests exceeds 300 μm. The data are then summarized to obtain the tool loss column vector u.

[0094] u=[u1,u2,u3,...,u i ,...,u m ] T

[0095] Among them, u i represents the damaged thickness of the PCD material layer on the flank face of the main cutting edge after the i-th group of cutting tests, m represents the total number of groups of single-factor cutting tests, and the value of m satisfies u m ≥300μm, and u m-1 <300μm.

[0096] 2.3) Time domain analysis is performed on the time domain working condition data set to obtain the time domain feature data set for different cutting lengths. The data attributes of each element in the data set include average value, rectified average value, maximum value, minimum value, peak value, valley value, peak-to-peak value, standard deviation, and root mean square value:

[0097]

[0098] in, represents the average value of the i-th group of time domain feature data sets, represents the rectified mean value of the i-th group of time domain feature data sets, represents the maximum value of the i-th group of time domain feature data sets, represents the minimum value of the i-th group of time domain feature data sets, represents the peak value of the i-th group of time domain feature data sets, represents the valley value of the i-th group of time domain feature data sets, represents the peak-to-peak value of the i-th group of time domain feature data sets, represents the standard deviation of the i-th group of time domain feature data sets, represents the root mean square of the time domain feature data set of group i, and m represents the total number of groups in the single-factor cutting test.

[0099] 2.4) Perform a discrete Fourier transform on the time domain working condition data set to obtain a frequency domain feature data set for different cutting lengths. The data attributes of each element in the data set include maximum amplitude, maximum amplitude frequency, center of gravity frequency, average amplitude, total energy in the frequency band, and the proportion of low-frequency energy between 40 and 200 Hz:

[0100]

[0101] in, represents the maximum amplitude of the i-th group of frequency domain feature data sets, Represents the maximum amplitude frequency of the i-th group of frequency domain feature data sets, represents the centroid frequency of the i-th group of frequency domain feature data sets, represents the average amplitude of the i-th group of frequency domain feature data sets, represents the total energy of the frequency band of the i-th group of frequency domain feature data sets, represents the proportion of low-frequency energy from 40 to 200 Hz in the i-th group of frequency domain feature data sets, and m represents the total number of groups in the single-factor cutting test.

[0102] 2.5) Combine the time domain feature data set and the frequency domain feature data set of the same cutting length to obtain a certain cutting length L i The row vector λ of the time-frequency domain characteristic data of the cutting condition at i :

[0103]

[0104] The time-frequency domain characteristic data row vectors of the working condition are further combined in the increasing order of cutting length to obtain the time-frequency domain characteristic data matrix D of the cutting condition.

[0105]

[0106] Among them, λ i Indicates cutting length L i The row vector of the time-frequency domain characteristic data of the cutting condition at time , λ i Transpose.

[0107] 2.6) Perform correlation analysis on each column vector in the cutting condition time-frequency domain feature data matrix D and the tool loss column vector u to obtain the reduced-dimensional cutting condition time-frequency domain feature data matrix D′, which contains Figure 2 The specific steps shown are:

[0108] 2.6.1) Calculate the Spearman correlation coefficient R between each column vector in the cutting condition time-frequency domain characteristic data matrix D and the tool loss column vector u in sequence. The calculation formula is:

[0109]

[0110] Among them, d i This represents the rank difference between the elements with the same index value in the two column vectors to be analyzed for Spearman correlation, sorted in descending order within the set of all elements in their respective column vectors. m represents the dimension of the column vector, which is also the total number of groups in the one-way cutting experiment.

[0111] The obtained correlation coefficient row vector r is:

[0112] r=[R1,R2,R3,...,R i ,...,R l ]T

[0113] Among them, R i It represents the Spearman correlation coefficient between the i-th column vector of the cutting condition time-frequency domain feature data matrix D and the tool loss column vector u, and l represents the number of columns in the matrix D.

[0114] 2.6.2) Construct a statistic S that follows the standard normal distribution N(0,1):

[0115]

[0116] Where R represents the Spearman correlation coefficient. m represents the number of samples, which in this embodiment is equivalent to the dimension of the column vector. s~N(0,1). Substitute the values ​​in the correlation coefficient row vector r into the calculation formula of the statistic S in turn to obtain the test value row vector s:

[0117] s=[S1,S2,S3,...,S i ,...,S l ]T

[0118] Among them, S i It represents the test value of the Spearman correlation analysis between the i-th column vector of the cutting condition time-frequency domain characteristic data matrix D and the tool loss column vector u, and l represents the number of columns in the matrix D.

[0119] 2.6.3) Using the obtained test value row vector s, calculate each P value in turn. The calculation formula is:

[0120]

[0121] Where S represents the test value. Substitute each test value in the obtained test value row vector s into the P value calculation formula to calculate the P value row vector p:

[0122] p=[P1,P2,P3,...,P i ,...,P l ]T

[0123] Among them, P i represents the P value of the Spearman correlation analysis between the i-th column vector of the cutting condition time-frequency domain feature data matrix D and the tool loss column vector u, and l represents the number of columns in the matrix D.

[0124] 2.6.4) Determine the significance level of the correlation between the corresponding column vectors of the cutting condition time-frequency domain characteristic data matrix D and the tool wear column vector u based on the P-value row vector p.

[0125] If P i ≤0.05, indicating that at the 95% confidence level, the i-th column vector of the cutting condition time-frequency domain characteristic data matrix D has a strong correlation with the tool loss column vector u.

[0126] If P i <0.05, indicating that at the 95% confidence level, the i-th column vector of the cutting condition time-frequency domain characteristic data matrix D has no strong correlation with the tool loss column vector u, and the column is marked as a redundant column vector.

[0127] 2.6.5) Remove redundant column vectors from the cutting condition time-frequency domain feature data matrix D to obtain the reduced-dimensional cutting condition time-frequency domain feature data matrix D′.

[0128] 2.7) Using the reduced-dimensional cutting condition time-frequency domain feature data matrix D′ as input and the tool loss column vector u as output, a tool loss prediction model based on a shallow neural network is established:

[0129] u=Φ(D′)

[0130] 3) Real-time acquisition of working condition signals during the cutting process, and calculation of the estimated PCD tool loss value using the established PCD tool loss value prediction model, specifically including the following steps:

[0131] 3.1) The radial vibration data of the tool during the cutting process is collected in real time at a certain sampling frequency f, and the data is aggregated at a given time interval Δt to obtain a segmented time domain working condition data set within a certain sampling interval time period t~t+Δt.

[0132]

[0133] in, It represents the radial vibration acceleration amplitude of the tool at the kth sampling moment in the sampling interval from t to t + Δt.

[0134] 3.2) Perform time domain analysis on the segmented time domain operating condition data set to obtain a segmented time domain feature data set. The data attributes of each element in the data set include average value, rectified average value, maximum value, minimum value, peak value, valley value, peak-to-peak value, standard deviation, and root mean square value:

[0135]

[0136] in, Represents the average value of the time domain feature data set within the sampling interval of t~t+Δt, Represents the rectified average value of the time domain feature data set within the sampling interval of t~t+Δt, Indicates the maximum value of the time domain feature data set within the sampling interval of t~t+Δt, Indicates the minimum value of the time domain feature data set within the sampling interval of t~t+Δt, Represents the peak value of the time domain feature data set within the sampling interval of t~t+Δt, Represents the valley value of the time domain feature data set within the sampling interval of t~t+Δt, Represents the peak-to-peak value of the time domain feature data set within the sampling interval of t~t+Δt, Represents the standard deviation of the time domain feature data set within the sampling interval of t~t+Δt, It represents the RMS value of the time domain feature dataset within the sampling interval from t to t+Δt.

[0137] 3.3) Perform discrete Fourier transform on the segmented time domain operating condition data set to obtain a segmented frequency domain feature data set. The data attributes of each element in the data set include maximum amplitude, maximum amplitude frequency, center of gravity frequency, average amplitude, total energy in the frequency band, and the proportion of low-frequency energy from 40 to 200 Hz:

[0138]

[0139] in, Indicates the maximum amplitude of the frequency domain feature data set within the sampling interval of t~t+Δt, Indicates the maximum amplitude frequency of the frequency domain feature data set within the sampling interval of t~t+Δt, Represents the centroid frequency of the frequency domain feature data set within the sampling interval t~t+Δt, Represents the average amplitude of the frequency domain feature data set within the sampling interval of t~t+Δt, Represents the total energy of the frequency band of the frequency domain feature data set within the sampling interval of t~t+Δt, Indicates the proportion of low-frequency energy between 40 and 200 Hz in the frequency domain feature dataset within the sampling interval from t to t+Δt.

[0140] 3.4) Combine the segmented time domain feature data set and the frequency domain feature data set to obtain the segmented time and frequency domain feature data row vector λ of the cutting condition t :

[0141]

[0142] 3.5) The row vector λ of the time-frequency domain characteristic data of the cutting condition segmentation t Perform dimensionality reduction and remove the row vector λ t The elements with the same column number as the redundant column vector removed in the dimensionality reduction process of the cutting condition time-frequency domain feature data matrix D in step S2 are obtained to obtain the row vector λ of the cutting condition segmented time-frequency domain feature data after dimensionality reduction t ′.

[0143] 3.6) The time-frequency domain feature row vector λ of the cutting condition segment after dimensionality reduction t ′ is input, and the tool loss estimation value u is calculated using the established tool loss value prediction model e :

[0144] u e =Φ(λ t ′)

[0145] 4) Determine the estimated loss value u of the PCD tool e and the preset loss threshold u b When the loss estimate of PCD tool u e Greater than the preset loss threshold u b When the cutting speed is 0.05, it is determined as an abnormal working condition of ceramic matrix composite material cutting; otherwise, it is determined as a normal working condition.

Claims

1. A real-time sensing method for abnormal cutting conditions of ceramic matrix composite materials based on PCD tools, characterized in that: The following steps are involved: 1) Carry out orthogonal cutting tests on ceramic matrix composites based on PCD tools to obtain the optimal cutting parameter combination {S, V, a w ,a p }; Where S represents the spindle speed, V represents the feed speed, a w Indicates the cutting width, a p Indicates cutting depth; Step 1) specifically includes the following sub-steps: 1.1) Conduct orthogonal cutting tests on ceramic matrix composites using PCD cutting tools. Collect radial vibration data of the cutting tool during each cutting test at a certain sampling frequency to obtain a time domain working condition dataset. 1.2) After each cutting test, the thickness of the PCD material layer on the flank surface of the tool's main cutting edge and secondary cutting edge was measured as the tool's main and secondary cutting edge loss values; 1.3) Conduct range analysis based on the primary and secondary cutting edge loss values ​​to obtain the cutting parameter combination based on the principle of minimum primary and secondary cutting edge loss values, including spindle speed, feed rate, cutting width and cutting depth; 2) Based on the cutting parameter combination obtained in step 1), a single-factor cutting test on cutting length was carried out to establish a tool loss value prediction model based on a shallow neural network: u=Φ(D′) Where D′ is the time-frequency domain characteristic data matrix of the cutting condition after dimension reduction, and u is the column vector of tool loss; 3) Real-time acquisition of working condition signals during the cutting process, and the use of the established tool loss value prediction model to calculate the tool loss estimate u e ; 4) Determine the estimated value of tool loss u e and the preset loss threshold u b The relationship between the size of the PCD tool and the abnormal working condition of the ceramic matrix composite material cutting is identified; among them, when the loss estimate value u e Greater than the preset loss threshold u b When the cutting speed is 0.05, it is determined as an abnormal working condition of ceramic matrix composite material cutting; otherwise, it is determined as a normal working condition.

2. The method for real-time sensing of abnormal cutting conditions of ceramic matrix composite materials based on PCD cutting tools according to claim 1, characterized in that: In step 2), a single-factor cutting test with equidistant increasing cutting length is carried out.

3. The method for real-time perception of abnormal working conditions in ceramic matrix composite cutting based on PCD cutting tools according to claim 1, characterized in that: Step 2) specifically includes the following sub-steps: 2.1) Based on the optimal cutting parameter combination obtained, single-factor cutting tests were conducted with equidistant increasing cutting lengths. During each cutting test, radial vibration data of the tool was collected at a certain sampling frequency to obtain a time-domain working condition dataset for different cutting lengths. 2.2) After each cutting test, the thickness of the material layer damaged on the flank face of the tool's main cutting edge is measured as the tool loss value for the current cutting length. A new tool is then replaced for the next test. This process continues until the tool loss value after a set of cutting tests exceeds 300 μm. The data is then summarized to obtain a tool loss column vector. 2.3) Perform time domain analysis on the time domain working condition data set to obtain the time domain characteristic data set of different cutting lengths. The data attributes of each element in the data set include average value, rectified average value, maximum value, minimum value, peak value, valley value, peak-to-peak value, standard deviation, and root mean square value; 2.4) Performing a discrete Fourier transform on the time-domain working condition data set to obtain a frequency-domain feature dataset for different cutting lengths. The data attributes of each element in the dataset include maximum amplitude, maximum amplitude frequency, center of gravity frequency, average amplitude, total energy in the frequency band, and the proportion of low-frequency energy between 40 and 200 Hz. 2.5) combining the time domain feature data set and the frequency domain feature data set for the same cutting length to obtain a row vector of the time-frequency domain feature data of the cutting condition for a certain cutting length, and combining the row vectors of the time-frequency domain feature data of the condition in increasing order of cutting length to obtain a matrix of the time-frequency domain feature data of the cutting condition; 2.6) Obtain a reduced-dimensionality cutting condition time-frequency domain feature data matrix by analyzing the correlation between each column vector in the cutting condition time-frequency domain feature data matrix and the tool wear column vector; 2.7) Using the reduced-dimensional cutting condition time-frequency domain feature data matrix as input and the tool loss column vector as output, a tool loss value prediction model based on a shallow neural network is established.

4. The method for real-time perception of abnormal working conditions in cutting ceramic matrix composite materials based on PCD tools according to claim 3 is characterized in that: Step 2.6) specifically includes the following sub-steps: 2.6.1) Calculate the Spearman correlation coefficient between each column vector of the cutting condition time-frequency domain characteristic data matrix and the tool wear column vector to obtain a row vector of correlation coefficients; 2.6.2) Construct a statistic that follows a standard normal distribution and, using the obtained Spearman correlation coefficient row vector, calculate each test value in turn to obtain a test value row vector; 2.6.3) Using the obtained test value row vector, calculate each P value in turn to obtain a P value row vector; 2.6.4) Determine the significance level of the correlation between the corresponding column vectors of the cutting condition time-frequency domain characteristic data matrix and the tool wear column vector based on the P-value row vectors, and mark the redundant column vectors; 2.6.5) Remove redundant column vectors from the cutting condition time-frequency domain feature data matrix to obtain a reduced-dimensional cutting condition time-frequency domain feature data matrix.

5. The method for real-time perception of abnormal working conditions in ceramic matrix composite cutting based on PCD cutting tools according to claim 4 is characterized in that: Step 3) specifically includes the following sub-steps: 3.1) Collect radial vibration data of the tool during the cutting process in real time at a certain sampling frequency, aggregate the data at given time intervals, and obtain a segmented time domain working condition data set within a certain sampling interval; 3.2) Perform time domain analysis on the segmented time domain operating condition data set to obtain a segmented time domain feature data set. The data attributes of each element in the data set include average value, rectified average value, maximum value, minimum value, peak value, valley value, peak-to-peak value, standard deviation, and root mean square value; 3.3) Perform discrete Fourier transform on the segmented time-domain operating condition dataset to obtain a segmented frequency-domain feature dataset. The data attributes of each element in the dataset include maximum amplitude, maximum amplitude frequency, center of gravity frequency, average amplitude, total energy in the frequency band, and the proportion of low-frequency energy from 40 to 200 Hz. 3.4) combining the segmented time domain feature data set and the frequency domain feature data set to obtain a row vector of the segmented time and frequency domain feature data of the cutting condition; 3.5) performing dimensionality reduction processing on the row vectors of the time-frequency domain characteristic data of the cutting condition segment, removing the elements in the row vectors that have the same column numbers as the redundant column vectors removed by the dimensionality reduction process of the time-frequency domain characteristic data matrix of the cutting condition in step 2.6); 3.6) Taking the time-frequency domain feature row vector of the cutting condition after dimensionality reduction as input, the tool loss estimation value is calculated using the established tool loss value prediction model.

Images