A small-diameter ball head grinding wheel wear state prediction method and system based on feature dimension reduction and radial basis neural network combination

Abstract

This invention discloses a method and system for predicting the wear state of small-diameter ball-end grinding wheels based on a combination of feature reduction and radial basis function neural networks. It relates to the field of ultra-precision machining technology and addresses the problem that existing methods often use single or overly complex wear state feature values, making it difficult to obtain accurate and stable prediction models. The method includes the following steps: S1, mounting an acoustic emission sensor array on the grinding wheel spindle mounting bracket of a grinding machine, collecting acoustic emission signals under different grinding wheel wear states using the acoustic emission sensors, and preprocessing the collected signals; S2, extracting feature values ​​related to the grinding wheel wear state from the acoustic emission signals, and performing feature reduction through correlation calculation; S3, constructing a grinding wheel wear state prediction model using a radial basis function neural network based on the reduced acoustic emission signal feature values ​​and corresponding wear states, establishing a mapping relationship between feature values ​​and wear states, and ultimately achieving prediction of the wear state of small-diameter ball-end grinding wheels under grinding conditions.

Description

Technical Field

[0001] This invention relates to the field of ultra-precision machining technology, and more specifically, to a method and system for predicting the wear state of small-diameter ball-end grinding wheels based on a combination of feature reduction and radial basis function neural networks. Background Technology

[0002] The hemispherical resonator gyroscope is a new type of solid-state resonator gyroscope characterized by high precision, high reliability, and long lifespan, and is therefore widely used in inertial systems such as aerospace and satellites. The hemispherical resonator is the core component of the hemispherical resonator gyroscope, and its machining quality and precision directly determine its performance. Because the resonator is made of high-purity fused silica, a typical difficult-to-machine material, wear of the grinding wheel is inevitable during the grinding process, thus reducing the machining quality and surface finish of the hemispherical resonator. Therefore, real-time monitoring and evaluation of the grinding wheel wear state is of great significance to the ultra-precision grinding process.

[0003] In a conventional sense, there is no strict definition of the severity of grinding wheel wear. It is typically categorized based on characteristics such as the form of abrasive grain shedding from the grinding wheel surface, grinding vibration signals, and the surface morphology of the grinding wheel. To better characterize the wear state of grinding wheels, offline detection and in-situ detection are two commonly used methods. Offline detection, based on measuring equipment, can clearly observe the surface morphology and abrasive grain shedding of the grinding wheel, but it negatively impacts grinding efficiency and accuracy. In-situ detection, on the other hand, uses various sensors such as acoustic emission sensors and vibration sensors to collect signal characteristics associated with the grinding wheel wear state. A mathematical model is then used to establish the relationship between these signal characteristics and the wear state. However, this detection technique requires complex logical operations and suffers from problems such as the difficulty in obtaining accurate and stable prediction models due to the selection of single or overly complex wear state characteristic values, unclear classification of grinding wheel wear states, significant influence of subjective human factors on the results, and cumbersome and inefficient modeling and analysis processes. Summary of the Invention

[0004] The technical problem to be solved by this invention is:

[0005] Existing methods select wear state characteristic values ​​that are either too simple or too complex, making it difficult to obtain accurate and stable prediction models. The resulting grinding wheel wear state classification is unclear, the results are greatly affected by subjective human factors, and the analysis process is cumbersome.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0007] This invention provides a method for predicting the wear state of small-diameter ball-end grinding wheels based on a combination of feature reduction and radial basis function neural networks, comprising the following steps:

[0008] S1. Install the acoustic emission sensor array on the grinding wheel spindle fixing frame of the grinding equipment, use the acoustic emission sensor to collect acoustic emission signals under different grinding wheel wear conditions, and preprocess the collected acoustic emission signals.

[0009] S2. Extract the feature values ​​related to the wear state of the grinding wheel from the acoustic emission signal, and perform feature dimensionality reduction through correlation calculation;

[0010] S3. Based on the dimensionality-reduced acoustic emission signal feature values ​​and the corresponding wear states, construct a radial basis neural network model for predicting the wear state of grinding wheels, establish the mapping relationship between feature values ​​and wear states, and finally realize the prediction of the wear state of small-diameter ball-end grinding wheels under grinding conditions.

[0011] Furthermore, the acoustic emission sensor described in S1 is an array of multiple sensors, and the array forms include: circular array, rectangular array and cross array.

[0012] Furthermore, the preprocessing of the acquired acoustic emission signal described in S1 includes filtering and denoising the acquired acoustic emission signal based on the Hilbert-Huang transform, and then normalizing the filtered and denoised acoustic emission signal.

[0013] Furthermore, the normalization process specifically includes:

[0014]

[0015] In the formula, x_GY i (t) represents the acoustic emission signal x sampled during the i-th grinding cycle. i The normalized result of x1(t) max With x1(t) min These represent the maximum and minimum values ​​of the sampled signal x1(t) during the first grinding cycle, respectively. i (t) represents the acoustic emission signal sampled during the i-th grinding cycle.

[0016] Furthermore, S2 includes the following steps:

[0017] Step 1: Extract feature values ​​related to the wear state of the grinding wheel from the acoustic emission signal and construct a feature set X = {X1, X2, X3, ... X}. M}, M represents the number of eigenvalues ​​in the acoustic emission signal related to the wear state of the grinding wheel. Calculate the correlation coefficient between eigenvalue X1 and the remaining eigenvalues ​​in the feature set X. If eigenvalue X1 exists... j The correlation coefficient between M and X1, j∈2,3,4… If the value is greater than a preset threshold, delete the feature value X. j Otherwise, keep X. jThis is to achieve dimensionality reduction of the feature set X;

[0018] Step 2: Calculate the correlation coefficient between the second feature value and the remaining feature values ​​in the dimensionality-reduced feature set, and further reduce the dimensionality of the feature set according to the method in Step 1; repeat the above process until the correlation calculation of all feature values ​​is completed, thus achieving feature dimensionality reduction.

[0019] Furthermore, the correlation coefficient mentioned in step 1 The calculation method is as follows:

[0020]

[0021] Wherein, parameter L represents the number of acoustic emission signals collected, x s Let y be the result of the feature value X1 in the s-th sampled signal. s Represented as eigenvalue X j The result in the s-th sampled signal, This represents the mean of the feature values ​​X1 among all sampled signals. X represents the characteristic value among all sampled signals. j The mean.

[0022] Furthermore, the eigenvalues ​​obtained after dimensionality reduction in step 2 include: mean. Peak factor C f Kurtosis index K and waveform factor W s .

[0023] Furthermore, the wear state of the grinding wheel described in S1 includes: no wear, initial wear, middle wear, and late wear, and the wear state of the grinding wheel is characterized by the number of grinding cycles.

[0024] Furthermore, the method of using the number of grinding cycles to characterize the wear state of the grinding wheel is as follows: a grinding cycle of 1 represents no wear, a grinding cycle of 2 to 10 represents the initial stage of wear, a grinding cycle of 11 to 100 represents the middle stage of wear, and a grinding cycle of 100 or more represents the later stage of wear.

[0025] A system for predicting the wear state of a small-diameter ball-end grinding wheel based on a combination of feature reduction and radial basis function neural network is provided. The system has a program module corresponding to the steps of any of the above-mentioned technical solutions, and executes the steps in the above-mentioned method for predicting the wear state of a small-diameter ball-end grinding wheel based on a combination of feature reduction and radial basis function neural network during operation.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] This invention discloses a method and system for predicting the wear state of small-diameter ball-end grinding wheels based on a combination of feature reduction and radial basis function neural networks. By using correlation calculation to perform feature reduction based on the feature values ​​in the acoustic emission signal that are related to the wear state of the grinding wheel, the resulting feature set exhibits low correlation between the various feature values. This allows for the establishment of a more accurate grinding wheel wear state prediction model using fewer signal feature values, reducing modeling complexity while ensuring the reliability of the prediction results. Ultimately, this enables online prediction of the wear state of small-diameter ball-end grinding wheels during ultra-precision grinding, providing technical guidance for timely wheel dressing or replacement, and laying the foundation for efficient and high-quality development of ultra-precision grinding.

[0028] The method of this invention has a certain degree of universality. It can not only be used to study the wear state of small-diameter ball-head grinding wheels during ultra-precision grinding of spherical parts, but can also be extended to the study of grinding wheel wear state during ultra-precision grinding of aspherical parts. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method for predicting the wear state of a small-diameter ball-end grinding wheel based on the combination of feature dimensionality reduction and radial basis neural network in an embodiment of the present invention;

[0030] Figure 2 This is a schematic diagram of the acoustic emission sensor sampling system in an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of an ultra-precision machine tool and an acoustic emission sensor array in an embodiment of the present invention;

[0032] Figure 4 This is a schematic diagram of the acoustic emission sensor array configuration in an embodiment of the present invention;

[0033] Figure 5 These are the time-domain signals of grinding processes under different grinding cycles in the embodiments of the present invention;

[0034] Figure 6 This is a comparison diagram of the signal before and after based on the Hilbert-Huang transform in an embodiment of the present invention;

[0035] Figure 7 This is a diagram of the radial basis function (RBF) neural network structure in an embodiment of the present invention;

[0036] Figure 8 This is a comparison chart of the grinding wheel wear theory and prediction results in the embodiments of the present invention.

[0037] Explanation of reference numerals in the attached figures:

[0038] 1-C-axis rotary table, 2-grinding wheel spindle holder, 3-grinding wheel spindle, 4-ball end grinding wheel, 5-workpiece to be processed, 6-machine tool horizontal worktable, 7-workpiece spindle, 8-workpiece spindle protective cover, 9-acoustic emission sensor. Detailed Implementation

[0039] In the description of this invention, it should be noted that the terms "first," "second," and "third" mentioned in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," and "third" may explicitly or implicitly include one or more of that feature.

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0041] Specific Implementation Plan 1: (e.g.) Figure 1 As shown, this invention provides a method for predicting the wear state of a small-diameter ball-end grinding wheel based on a combination of feature reduction and radial basis function neural network, comprising the following steps:

[0042] S1, such as Figure 2 As shown, an acoustic emission sensor array is installed on the grinding wheel spindle mounting bracket of a grinding equipment. The acoustic emission sensors are used to collect acoustic emission signals under different grinding wheel wear conditions, and the collected acoustic emission signals are preprocessed.

[0043] S2. Extract the feature values ​​related to the wear state of the grinding wheel from the acoustic emission signal, and perform feature dimensionality reduction through correlation calculation;

[0044] S3. Based on the dimensionality-reduced acoustic emission signal feature values ​​and the corresponding wear states, construct a radial basis neural network model for predicting the wear state of grinding wheels, establish the mapping relationship between feature values ​​and wear states, and finally realize the prediction of the wear state of small-diameter ball-end grinding wheels under grinding conditions.

[0045] In this implementation plan, a four-axis, three-linkage ultra-precision grinding and polishing machine is used to grind small-diameter ball-end grinding wheels and complex thin-walled components. It features three linear motion axes (X, Y, and Z axes) and a C-axis rotary table 1. During grinding, the relative grinding speed between the workpiece 5 and the ball-end grinding wheel 4 is adjusted by changing the rotational speeds of the workpiece spindle 7 and the grinding wheel spindle 3. The feed rate is adjusted by controlling the movement speed of the machine tool's horizontal worktable 6. Because the grinding fluid is constantly sprayed onto the contact area between the ball-end grinding wheel 4 and the workpiece 5 during ultra-precision grinding, such as... Figure 3 As shown, in order to avoid the influence of continuous impact of grinding fluid on the signal, the acoustic emission sensor 9 is fixed to the grinding wheel spindle holder 2, which is not affected by the grinding fluid, using a coupling agent array.

[0046] Since the grinding cycle T≈30min in ultra-precision grinding is too long to collect all the data for the entire grinding cycle, a segment of signal is extracted from the grinding cycle for analysis, and it is ensured that each sampled data is in the same grinding area.

[0047] like Figure 5 As shown, in order to obtain acoustic emission signals under different wear conditions, acoustic emission signals generated during 1 cycle, 3 cycles, 5 cycles, ..., N cycles of ball head grinding were collected.

[0048] Specific Implementation Plan Two: (e.g.) Figure 4 As shown, the acoustic emission sensor in S1 is an array of multiple sensors, and the array forms include: circular array, rectangular array, and cross array. Other aspects of this embodiment are the same as in specific embodiment one.

[0049] Since the more training samples a radial basis function neural network has, the higher the accuracy of subsequent grinding wheel wear state prediction results, an array of multiple sensors is used for signal acquisition. Considering the space and number of sensors, it is preferable to arrange the acoustic emission sensors in a rectangular array.

[0050] Specific Implementation Scheme 3: The preprocessing of the acquired acoustic emission signal described in S1 includes filtering and denoising the acquired acoustic emission signal based on the Hilbert-Huang transform, and then normalizing the filtered and denoised acoustic emission signal. All other aspects of this implementation scheme are the same as in Specific Implementation Scheme 1.

[0051] Due to the continuous impact of grinding fluid, the surrounding environment, and machine tool vibration, the acoustic emission sensor's sampled signal contains interference signals. Directly extracting signal features from the raw data and establishing a grinding wheel wear prediction model would result in a significant deviation between the prediction and the actual values. Since acoustic emission signals are typically nonlinear and non-stationary, Fourier transform and wavelet transform have limitations in processing such signals. The Hilbert-Huang transform, however, can maximize signal noise reduction while preserving the basic signal characteristics. Therefore, this implementation scheme uses the Hilbert-Huang transform to achieve filtering and noise reduction of the sampled signal, as shown in the following figure. Figure 6 As shown.

[0052] Specific implementation plan four: The normalization process is as follows:

[0053]

[0054] In the formula, x_GY i (t) represents the acoustic emission signal x sampled during the i-th grinding cycle. i The normalized result of x1(t) maxWith x1(t) min These represent the maximum and minimum values ​​of the sampled signal x1(t) during the first grinding cycle, respectively. i (t) represents the acoustic emission signal sampled during the i-th grinding cycle. This implementation scheme is otherwise identical to specific implementation scheme three.

[0055] Specific implementation plan five: S2 includes the following steps:

[0056] Step 1: Extract feature values ​​related to the wear state of the grinding wheel from the acoustic emission signal and construct a feature set X = {X1, X2, X3, ... X}. M}, M represents the number of eigenvalues ​​in the acoustic emission signal related to the wear state of the grinding wheel. Calculate the correlation coefficient between eigenvalue X1 and the remaining eigenvalues ​​in the feature set X. If eigenvalue X1 exists... j The correlation coefficient between M and X1, j∈2,3,4… If the value is greater than a preset threshold, delete the feature value X. j Otherwise, keep X. j This is to achieve dimensionality reduction of the feature set X;

[0057] Step 2: Calculate the correlation coefficient between the second eigenvalue and the remaining eigenvalues ​​in the dimensionality-reduced feature set. Further reduce the dimensionality of the feature set using the method in Step 1. Repeat the above process until all eigenvalues ​​have completed their correlation calculations, thus achieving dimensionality reduction. This implementation scheme is otherwise identical to Implementation Scheme 1.

[0058] The preset threshold in this implementation scheme is 0.8.

[0059] Common characteristic values ​​include: peak value X p mean Root mean square value X rms Peak factor C f Kurtosis index K, waveform factor W s It has more than ten characteristic values, including impulse factor and margin factor. The expressions for some of the parameters are shown below:

[0060]

[0061]

[0062]

[0063]

[0064]

[0065] The parameter n is the number of data points sampled by the acoustic emission sensor.

[0066] Since there are many signal feature values, when using radial basis neural networks to build a prediction model, the selected signal feature set needs to be dimensionality reduced. The correlation between the feature values ​​in the dimensionality-reduced feature set is low, which can enable the establishment of a more accurate grinding wheel wear state prediction model using fewer signal feature values.

[0067] Specific Implementation Plan Six: The Correlation Coefficients Mentioned in Step 1 The calculation method is as follows:

[0068]

[0069] Wherein, parameter L represents the number of acoustic emission signals collected, x s Let y be the result of the feature value X1 in the s-th sampled signal. s Represented as eigenvalue X j The result in the s-th sampled signal, This represents the mean of the feature values ​​X1 among all sampled signals. X represents the characteristic value among all sampled signals. j The average value. This implementation plan is otherwise the same as Implementation Plan Five.

[0070] Specific implementation plan seven: The feature values ​​obtained after feature dimensionality reduction in step 2 include: mean Peak factor C f Kurtosis index K and waveform factor W s This implementation plan is otherwise the same as specific implementation plan six.

[0071] Specific Implementation Scheme Eight: The grinding wheel wear state described in S1 includes: no wear, initial wear, intermediate wear, and late wear, and the grinding wheel wear state is characterized by the number of grinding cycles. All other aspects of this implementation scheme are the same as Specific Implementation Scheme One.

[0072] Specific Implementation Scheme Nine: The method of characterizing the wear state of the grinding wheel using the number of grinding cycles is as follows: 1 grinding cycle represents no wear, 2-10 grinding cycles represent the initial stage of wear, 11-100 grinding cycles represent the middle stage of wear, and 100 or more grinding cycles represent the late stage of wear. All other aspects of this implementation scheme are the same as in Specific Implementation Scheme Eight.

[0073] Since small-diameter ball-end grinding wheels undergo four stages from installation to replacement—namely, no wear (initial grinding), initial wear, intermediate wear, and late wear—and generally, the larger the number of grinding cycles T, the more severe the wear of the grinding wheel, the number of ultra-precision grinding cycles T is used to characterize the wear state of small-diameter ball-end grinding wheels. Table 1 shows the acoustic emission signal characteristic values ​​after dimensionality reduction of some features and the corresponding wear state data.

[0074] Table 1

[0075]

[0076] Specific Implementation Scheme 10: A small-diameter ball-end grinding wheel wear state prediction system based on feature reduction and radial basis function neural network. The system has a program module corresponding to the steps of any of the above implementation schemes. When running, it executes the steps in the above-mentioned small-diameter ball-end grinding wheel wear state prediction method based on feature reduction and radial basis function neural network.

[0077] Example 1

[0078] To verify the accuracy of the method of this invention, acoustic emission signals generated during 1, 3, 5, ..., 121 cycles of ball-end grinding were collected. The method of this invention was used to preprocess the data and perform feature reduction, such as... Figure 7 As shown, a radial basis function neural network (RBF) model for predicting grinding wheel wear was constructed and trained. Acoustic emission signals were collected using an acoustic emission sensor for grinding cycles T of 1, 3, 8, 15, 50, 80, 110, and 150, and the corresponding feature values ​​were extracted and substituted into the trained model as input variables. The results of the grinding wheel wear state are shown in Table 3. Figure 8 As shown.

[0079] Table 2

[0080]

[0081] According to Table 2 and Figure 8 (a) It can be seen that although there is a certain amount of error between the grinding wheel wear degree predicted by the radial basis function neural network and the theoretical wear state of the grinding wheel, the reason for this error is the insufficient accuracy of the model due to the small number of training samples. However, it has achieved accurate prediction of the grinding wheel wear state. Figure 8 As shown in (b), the curve exhibits a trend of first decreasing and then increasing, indicating that the prediction accuracy for no wear, early wear stage, and mid-wear stage is better than that for late wear stage. In practical applications, it is necessary to ensure that the ball-end grinding wheel is in the early to mid-wear stage; therefore, the method of this invention has good effectiveness and practicality.

[0082] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A small-diameter ball head grinding wheel wear state prediction method based on feature dimension reduction combined with a radial basis neural network, characterized in that, Includes the following steps: S1. Install the acoustic emission sensor array on the grinding wheel spindle fixing frame of the grinding equipment, use the acoustic emission sensor to collect acoustic emission signals under different grinding wheel wear conditions, and preprocess the collected acoustic emission signals. S2. Extract the feature values ​​related to the wear state of the grinding wheel from the acoustic emission signal, and perform feature dimensionality reduction through correlation calculation; S3. Based on the dimensionality-reduced acoustic emission signal feature values ​​and the corresponding wear states, construct a radial basis neural network model for predicting the wear state of grinding wheels, establish the mapping relationship between feature values ​​and wear states, and finally realize the prediction of the wear state of small-diameter ball-end grinding wheels under grinding conditions. The acoustic emission sensor described in S1 is an array of multiple sensors, and the array forms include: circular array, rectangular array and cross array; S2 includes the following steps: Step 1: Extract feature values ​​related to the wear state of the grinding wheel from the acoustic emission signal to construct a feature set. , This represents the number of eigenvalues ​​in the acoustic emission signal that are related to the wear state of the grinding wheel, and the eigenvalues ​​are calculated. With feature set The correlation coefficients between the remaining eigenvalues, if eigenvalues ​​exist , and Correlation coefficient between If the value exceeds a preset threshold, delete the feature value. Otherwise, keep To achieve the feature set Dimensionality reduction; Step 2: Calculate the correlation coefficient between the second feature value and the remaining feature values ​​in the dimensionality-reduced feature set, and further reduce the dimensionality of the feature set according to the method in Step 1; repeat the above process until the correlation calculation of all feature values ​​is completed, thus achieving feature dimensionality reduction; The correlation coefficient mentioned in step 1 The calculation method is as follows: Among them, parameters This indicates the number of acoustic emission signals collected. Represented as eigenvalues In the The result in each sampled signal Represented as eigenvalues In the The result in each sampled signal Represents the characteristic values ​​in all sampled signals The mean, Represents the characteristic values ​​in all sampled signals The mean; The wear state of the grinding wheel described in S1 includes: no wear, initial wear, middle wear, and late wear, and the number of grinding cycles is used to characterize the wear state of the grinding wheel; The grinding wheel wear state is characterized by the number of grinding cycles, specifically: 1 grinding cycle represents no wear, 2 to 10 grinding cycles represent the initial stage of wear, 11 to 100 grinding cycles represent the middle stage of wear, and 100 or more grinding cycles represent the later stage of wear.

2. The method for predicting the wear state of a small-diameter ball-end grinding wheel based on feature reduction and radial basis function neural network as described in claim 1, is characterized in that, The preprocessing of the acquired acoustic emission signal described in S1 includes filtering and denoising the acquired acoustic emission signal based on the Hilbert-Huang transform, and then normalizing the filtered and denoised acoustic emission signal.

3. The method for predicting the wear state of a small-diameter ball-end grinding wheel based on feature reduction and radial basis function neural network as described in claim 2, is characterized in that... The normalization process specifically includes: ， In the formula, The acoustic emission signal sampled during the i-th grinding cycle The normalization result, and These are the sampled signals during the first grinding cycle. The maximum and minimum values ​​in The acoustic emission signal sampled during the i-th grinding cycle.

4. The method for predicting the wear state of a small-diameter ball-end grinding wheel based on feature reduction and radial basis function neural network as described in claim 3, is characterized in that... The feature values ​​obtained after feature dimensionality reduction in step 2 include: mean. Peak factor kurtosis index and waveform factor .

5. A system for predicting the wear condition of a small-diameter ball-end grinding wheel based on a combination of feature reduction and radial basis function neural networks, characterized in that, The system has a program module corresponding to the steps of any one of the claims 1 to 4 above, and executes the steps in the above-described method for predicting the wear state of small-diameter ball-head grinding wheels based on the combination of feature dimensionality reduction and radial basis neural network when it is run.

Images